In [79]:

# Libraries
import numpy as np
import pandas as pd
pd.set_option('max_columns', None)
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline

import datetime
import lightgbm as lgb
from scipy import stats
from sklearn.model_selection import train_test_split, StratifiedKFold, KFold, cross_val_score, GridSearchCV, RepeatedStratifiedKFold
from sklearn.preprocessing import StandardScaler
import os
import plotly.offline as py
py.init_notebook_mode(connected=True)
import plotly.graph_objs as go
import plotly.tools as tls
import xgboost as xgb
import lightgbm as lgb
from sklearn import model_selection
from sklearn.metrics import accuracy_score, roc_auc_score
from sklearn import metrics
import json
import ast
import time
from sklearn import linear_model
import eli5
from eli5.sklearn import PermutationImportance
import shap
from tqdm import tqdm_notebook
from mlxtend.feature_selection import SequentialFeatureSelector as SFS
from mlxtend.plotting import plot_sequential_feature_selection as plot_sfs
from sklearn.neighbors import NearestNeighbors
from sklearn.feature_selection import GenericUnivariateSelect, SelectPercentile, SelectKBest, f_classif, mutual_info_classif, RFE
import statsmodels.api as sm
import warnings
warnings.filterwarnings('ignore')
from catboost import CatBoostClassifier

# import json
import ujson as json
import altair as alt
from  altair import v3
from IPython.display import HTML

from plotly import tools
import plotly.graph_objs as go
from plotly.offline import init_notebook_mode, iplot
import warnings
warnings.filterwarnings("ignore")
init_notebook_mode(connected=True)

Data overview¶

In [81]:

train = pd.read_csv('/Users/mant1/input/dota_csv/train_features.csv', index_col='match_id_hash', engine='python')
target = pd.read_csv('/Users/mant1/input/dota_csv/train_targets.csv', index_col='match_id_hash', engine='python')
test = pd.read_csv('/Users/mant1/input/dota_csv/test_features.csv', index_col='match_id_hash', engine='python')
PATH = '/Users/mant1/input/dota_csv/'

In [82]:

target['next_roshan_team'].value_counts()

Out[82]:

Radiant    13190
Dire       12135
Name: next_roshan_team, dtype: int64

In [85]:

target.head()

Out[85]:

	game_time	radiant_win	duration	time_remaining	next_roshan_team
match_id_hash
a400b8f29dece5f4d266f49f1ae2e98a	155	False	992	837	NaN
b9c57c450ce74a2af79c9ce96fac144d	658	True	1154	496	NaN
6db558535151ea18ca70a6892197db41	21	True	1503	1482	Radiant
46a0ddce8f7ed2a8d9bd5edcbb925682	576	True	1952	1376	NaN
b1b35ff97723d9b7ade1c9c3cf48f770	453	False	2001	1548	NaN

In [86]:

train.head()

Out[86]:

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_kills	r1_deaths	r1_assists	r1_denies	r1_gold	r1_lh	r1_xp	r1_health	r1_max_health	r1_max_mana	r1_level	r1_x	r1_y	r1_stuns	r1_creeps_stacked	r1_camps_stacked	r1_rune_pickups	r1_firstblood_claimed	r1_teamfight_participation	r1_towers_killed	r1_roshans_killed	r1_obs_placed	r1_sen_placed	r2_hero_id	r2_kills	r2_deaths	r2_assists	r2_denies	r2_gold	r2_lh	r2_xp	r2_health	r2_max_health	r2_max_mana	r2_level	r2_x	r2_y	r2_stuns	r2_creeps_stacked	r2_camps_stacked	r2_rune_pickups	r2_firstblood_claimed	r2_teamfight_participation	r2_towers_killed	r2_roshans_killed	r2_obs_placed	r2_sen_placed	r3_hero_id	r3_kills	r3_deaths	r3_assists	r3_denies	r3_gold	r3_lh	r3_xp	r3_health	r3_max_health	r3_max_mana	r3_level	r3_x	r3_y	r3_stuns	r3_creeps_stacked	r3_camps_stacked	r3_rune_pickups	r3_firstblood_claimed	r3_teamfight_participation	r3_towers_killed	r3_roshans_killed	r3_obs_placed	r3_sen_placed	r4_hero_id	r4_kills	r4_deaths	r4_assists	r4_denies	r4_gold	r4_lh	r4_xp	r4_health	r4_max_health	r4_max_mana	r4_level	r4_x	r4_y	r4_stuns	r4_creeps_stacked	r4_camps_stacked	r4_rune_pickups	r4_firstblood_claimed	r4_teamfight_participation	r4_towers_killed	r4_roshans_killed	r4_obs_placed	r4_sen_placed	r5_hero_id	r5_kills	r5_deaths	r5_assists	r5_denies	r5_gold	r5_lh	r5_xp	r5_health	r5_max_health	r5_max_mana	r5_level	r5_x	r5_y	r5_stuns	r5_creeps_stacked	r5_camps_stacked	r5_rune_pickups	r5_firstblood_claimed	r5_teamfight_participation	r5_towers_killed	r5_roshans_killed	r5_obs_placed	r5_sen_placed	d1_hero_id	d1_kills	d1_deaths	d1_assists	d1_denies	d1_gold	d1_lh	d1_xp	d1_health	d1_max_health	d1_max_mana	d1_level	d1_x	d1_y	d1_stuns	d1_creeps_stacked	d1_camps_stacked	d1_rune_pickups	d1_firstblood_claimed	d1_teamfight_participation	d1_towers_killed	d1_roshans_killed	d1_obs_placed	d1_sen_placed	d2_hero_id	d2_kills	d2_deaths	d2_assists	d2_denies	d2_gold	d2_lh	d2_xp	d2_health	d2_max_health	d2_max_mana	d2_level	d2_x	d2_y	d2_stuns	d2_creeps_stacked	d2_camps_stacked	d2_rune_pickups	d2_firstblood_claimed	d2_teamfight_participation	d2_towers_killed	d2_roshans_killed	d2_obs_placed	d2_sen_placed	d3_hero_id	d3_kills	d3_deaths	d3_assists	d3_denies	d3_gold	d3_lh	d3_xp	d3_health	d3_max_health	d3_max_mana	d3_level	d3_x	d3_y	d3_stuns	d3_creeps_stacked	d3_camps_stacked	d3_rune_pickups	d3_firstblood_claimed	d3_teamfight_participation	d3_towers_killed	d3_roshans_killed	d3_obs_placed	d3_sen_placed	d4_hero_id	d4_kills	d4_deaths	d4_assists	d4_denies	d4_gold	d4_lh	d4_xp	d4_health	d4_max_health	d4_max_mana	d4_level	d4_x	d4_y	d4_stuns	d4_creeps_stacked	d4_camps_stacked	d4_rune_pickups	d4_firstblood_claimed	d4_teamfight_participation	d4_towers_killed	d4_roshans_killed	d4_obs_placed	d4_sen_placed	d5_hero_id	d5_kills	d5_deaths	d5_assists	d5_denies	d5_gold	d5_lh	d5_xp	d5_health	d5_max_health	d5_max_mana	d5_level	d5_x	d5_y	d5_stuns	d5_creeps_stacked	d5_camps_stacked	d5_rune_pickups	d5_firstblood_claimed	d5_teamfight_participation	d5_towers_killed	d5_roshans_killed	d5_obs_placed	d5_sen_placed
match_id_hash
a400b8f29dece5f4d266f49f1ae2e98a	155	22	7	1	11	11	0	0	0	0	543	7	533	358	600	350.93784	2	116	122	0.000000	0	0	1	0	0.000000	0	0	0	0	78	0	0	0	3	399	4	478	636	720	254.93774	2	124	126	0.000000	0	0	0	0	0.000000	0	0	0	0	14	0	1	0	0	304	0	130	700	700	242.93773	1	70	156	0.000000	0	0	1	0	0.000000	0	0	0	0	59	0	0	0	1	389	4	506	399	700	326.93780	2	170	86	0.000000	0	0	0	0	0.000000	0	0	0	0	77	0	0	0	0	402	10	344	422	800	314.93780	2	120	100	0.000000	0	0	0	0	0.000000	0	0	0	0	12	0	0	1	13	982	12	780	650	720	386.93787	3	82	170	0.000000	0	0	1	0	1.00	0	0	0	0	21	0	0	0	6	788	9	706	640	640	422.93790	3	174	90	0.000000	0	0	2	0	0.00	0	0	0	0	60	0	0	0	1	531	0	307	720	720	242.93773	2	180	84	0.299948	0	0	2	0	0.00	0	0	0	0	84	1	0	0	0	796	0	421	760	760	326.93780	2	90	150	0.000000	0	0	2	1	1.0	0	0	1	0	34	0	0	0	0	851	11	870	593	680	566.93805	3	128	128	0.000000	0	0	0	0	0.00	0	0	0	0
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	15	7	2	0	7	5257	52	3937	1160	1160	566.93805	8	76	78	0.000000	0	0	0	0	0.437500	0	0	0	0	96	3	1	2	3	3394	19	3897	1352	1380	386.93787	8	78	166	8.397949	0	0	4	0	0.312500	0	0	0	0	27	1	1	4	2	2212	4	2561	710	860	530.93800	6	156	146	11.964951	2	1	4	0	0.312500	0	0	3	1	63	4	0	3	12	4206	38	4459	420	880	482.93796	9	154	148	0.000000	0	0	3	0	0.437500	0	0	1	2	89	1	0	5	4	3103	14	2712	856	900	446.93793	6	150	148	21.697395	0	0	2	0	0.375000	1	0	0	0	58	1	2	0	4	2823	24	3281	700	700	686.93820	7	88	170	3.165901	1	1	3	0	0.25	0	0	1	0	14	1	6	0	1	2466	17	2360	758	1040	326.93780	6	156	98	0.066650	0	0	1	1	0.25	0	0	4	2	1	1	3	1	7	3624	29	3418	485	800	350.93784	7	124	144	0.299955	2	1	4	0	0.50	0	0	0	0	56	0	3	2	3	2808	18	2730	567	1160	410.93790	6	124	142	0.000000	0	0	6	0	0.5	0	0	0	0	92	0	2	0	1	1423	8	1136	800	800	446.93793	4	180	176	0.000000	0	0	0	0	0.00	0	0	0	0
6db558535151ea18ca70a6892197db41	21	23	0	0	0	101	0	0	0	0	176	0	0	680	680	506.93800	1	118	118	0.000000	0	0	0	0	0.000000	0	0	0	0	51	0	0	0	0	176	0	0	720	720	278.93777	1	156	104	0.000000	0	0	0	0	0.000000	0	0	0	0	44	0	0	0	0	176	0	0	568	600	254.93774	1	78	144	0.000000	0	0	1	0	0.000000	0	0	0	0	49	0	0	0	0	176	0	0	580	580	254.93774	1	150	78	0.000000	0	0	1	0	0.000000	0	0	0	0	53	0	0	0	0	176	0	0	580	580	374.93787	1	78	142	0.000000	0	0	1	0	0.000000	0	0	0	0	18	0	0	0	0	96	0	0	660	660	266.93774	1	180	178	0.000000	0	0	0	0	0.00	0	0	0	0	67	0	0	0	0	96	0	0	586	620	278.93777	1	100	174	0.000000	0	0	0	0	0.00	0	0	0	0	47	0	0	0	0	96	0	0	660	660	290.93777	1	178	112	0.000000	0	0	1	0	0.00	0	0	0	0	40	0	0	0	0	96	0	0	600	600	302.93777	1	176	110	0.000000	0	0	0	0	0.0	0	0	0	0	17	0	0	0	0	96	0	0	640	640	446.93793	1	162	162	0.000000	0	0	0	0	0.00	0	0	0	0
46a0ddce8f7ed2a8d9bd5edcbb925682	576	22	7	1	4	14	1	0	3	1	1613	0	1471	900	900	290.93777	4	170	96	2.366089	0	0	5	0	0.571429	0	0	0	0	99	1	0	1	2	2816	30	3602	878	1100	494.93796	8	82	154	0.000000	0	0	1	0	0.285714	0	0	0	0	101	3	1	1	9	4017	44	4811	980	980	902.93835	9	126	128	0.000000	0	0	2	1	0.571429	0	0	2	0	26	1	1	2	1	1558	2	1228	640	640	422.93790	4	120	138	7.098264	0	0	5	0	0.428571	0	0	2	0	41	0	0	1	30	3344	55	3551	1079	1100	362.93784	7	176	94	1.932884	0	0	0	0	0.142857	0	0	0	0	18	0	0	0	0	2712	69	2503	825	1160	338.93784	6	94	158	0.000000	3	1	4	0	0.00	0	0	0	0	98	1	3	0	5	2217	23	3310	735	880	506.93800	7	126	142	0.000000	0	0	1	0	0.50	0	0	1	0	8	0	1	1	6	3035	44	2508	817	860	350.93784	6	78	160	0.000000	0	0	1	0	0.50	0	0	0	0	69	0	2	0	0	2004	16	1644	1160	1160	386.93787	4	176	100	4.998863	0	0	2	0	0.0	0	0	0	0	86	0	1	0	1	1333	2	1878	630	740	518.93800	5	82	160	8.664527	3	1	3	0	0.00	0	0	2	0
b1b35ff97723d9b7ade1c9c3cf48f770	453	22	7	1	3	42	0	1	1	0	1404	9	1351	1000	1000	338.93784	4	80	164	9.930903	0	0	4	0	0.500000	0	0	0	0	69	1	0	0	0	1840	14	1693	868	1000	350.93784	5	78	166	1.832892	0	0	0	1	0.500000	0	0	0	0	27	0	1	0	0	1204	10	3210	578	860	792.93823	7	120	122	3.499146	0	0	0	0	0.000000	0	0	0	0	104	0	0	2	0	1724	21	1964	777	980	434.93793	5	138	94	0.000000	0	0	1	0	1.000000	0	0	0	0	65	1	2	0	0	1907	8	1544	281	820	446.93793	4	174	100	0.000000	0	0	6	0	0.500000	0	0	0	0	23	1	0	0	0	1422	10	1933	709	940	362.93784	5	84	170	11.030720	0	0	1	0	0.25	0	0	0	0	22	1	0	0	1	1457	12	1759	712	820	482.93796	5	174	106	2.199399	0	0	1	0	0.25	0	0	0	0	35	0	0	1	2	2402	35	3544	349	720	434.93793	7	128	126	0.000000	0	0	2	0	0.25	0	0	0	0	72	2	1	0	0	1697	12	1651	680	680	374.93787	4	176	108	13.596678	0	0	2	0	0.5	0	0	0	0	1	0	1	1	8	2199	32	1919	692	740	302.93777	5	104	162	0.000000	2	1	2	0	0.25	0	0	0	0

In [87]:

n_fold = 5
folds = StratifiedKFold(n_splits=n_fold, shuffle=True, random_state=42)

Training function

In [88]:

def train_model(X, X_test, y, params, folds, model_type='lgb', plot_feature_importance=False, averaging='usual', model=None):
    oof = np.zeros(len(X))
    prediction = np.zeros(len(X_test))
    scores = []
    feature_importance = pd.DataFrame()
    for fold_n, (train_index, valid_index) in enumerate(folds.split(X, y)):
        print('Fold', fold_n, 'started at', time.ctime())
        X_train, X_valid = X.loc[train_index], X.loc[valid_index]
        y_train, y_valid = y[train_index], y[valid_index]
        
        if model_type == 'lgb':
            train_data = lgb.Dataset(X_train, label=y_train)
            valid_data = lgb.Dataset(X_valid, label=y_valid)
            
            model = lgb.train(params,
                    train_data,
                    num_boost_round=20000,
                    valid_sets = [train_data, valid_data],
                    verbose_eval=1000,
                    early_stopping_rounds = 200)
            
            y_pred_valid = model.predict(X_valid)
            y_pred = model.predict(X_test, num_iteration=model.best_iteration)
            
        if model_type == 'xgb':
            train_data = xgb.DMatrix(data=X_train, label=y_train, feature_names=X_train.columns)
            valid_data = xgb.DMatrix(data=X_valid, label=y_valid, feature_names=X_train.columns)

            watchlist = [(train_data, 'train'), (valid_data, 'valid_data')]
            model = xgb.train(dtrain=train_data, num_boost_round=20000, evals=watchlist, early_stopping_rounds=200, verbose_eval=500, params=params)
            y_pred_valid = model.predict(xgb.DMatrix(X_valid, feature_names=X_train.columns), ntree_limit=model.best_ntree_limit)
            y_pred = model.predict(xgb.DMatrix(X_test, feature_names=X_train.columns), ntree_limit=model.best_ntree_limit)
        
        if model_type == 'sklearn':
            model = model
            model.fit(X_train, y_train)
            y_pred_valid = model.predict_proba(X_valid).reshape(-1,)
            score = roc_auc_score(y_valid, y_pred_valid)
            # print(f'Fold {fold_n}. AUC: {score:.4f}.')
            # print('')
            
            y_pred = model.predict_proba(X_test)[:, 1]
            
        if model_type == 'glm':
            model = sm.GLM(y_train, X_train, family=sm.families.Binomial())
            model_results = model.fit()
            model_results.predict(X_test)
            y_pred_valid = model_results.predict(X_valid).reshape(-1,)
            score = roc_auc_score(y_valid, y_pred_valid)
            
            y_pred = model_results.predict(X_test)
            
        if model_type == 'cat':
            model = CatBoostClassifier(iterations=20000, learning_rate=0.05, loss_function='Logloss',  eval_metric='AUC', **params)
            model.fit(X_train, y_train, eval_set=(X_valid, y_valid), cat_features=[], use_best_model=True, verbose=False)

            y_pred_valid = model.predict_proba(X_valid)[:, 1]
            y_pred = model.predict_proba(X_test)[:, 1]
            
        oof[valid_index] = y_pred_valid.reshape(-1,)
        scores.append(roc_auc_score(y_valid, y_pred_valid))

        if averaging == 'usual':
            prediction += y_pred
        elif averaging == 'rank':
            prediction += pd.Series(y_pred).rank().values  
        
        if model_type == 'lgb':
            # feature importance
            fold_importance = pd.DataFrame()
            fold_importance["feature"] = X.columns
            fold_importance["importance"] = model.feature_importance()
            fold_importance["fold"] = fold_n + 1
            feature_importance = pd.concat([feature_importance, fold_importance], axis=0)

    prediction /= n_fold
    
    print('CV mean score: {0:.4f}, std: {1:.4f}.'.format(np.mean(scores), np.std(scores)))
    
    if model_type == 'lgb':
        feature_importance["importance"] /= n_fold
        if plot_feature_importance:
            cols = feature_importance[["feature", "importance"]].groupby("feature").mean().sort_values(
                by="importance", ascending=False)[:50].index

            best_features = feature_importance.loc[feature_importance.feature.isin(cols)]

            plt.figure(figsize=(16, 12));
            sns.barplot(x="importance", y="feature", data=best_features.sort_values(by="importance", ascending=False));
            plt.title('LGB Features (avg over folds)');
        
            return oof, prediction, feature_importance
        return oof, prediction, scores
    
    else:
        return oof, prediction, scores

In [89]:

param = {
        'bagging_freq': 5,  #handling overfitting
        'bagging_fraction': 0.5,  #handling overfitting - adding some noise
        'boost_from_average':'false',
        'boost': 'gbdt',
        'feature_fraction': 0.05, #handling overfitting
        'learning_rate': 0.01,  #the changes between one auc and a better one gets really small thus a small learning rate performs better
        'max_depth': -1,  
        'metric':'auc',
        'min_data_in_leaf': 50,
        'min_sum_hessian_in_leaf': 10.0,
        'num_leaves': 10,
        'num_threads': 5,
        'tree_learner': 'serial',
        'objective': 'binary', 
        'verbosity': 1,
        'seed': 17
    }

In [90]:

params2 = {
       'bagging_freq': 5,  #handling overfitting
       'bagging_fraction': 0.5,  #handling overfitting - adding some noise
       'boost_from_average':'false',
       'boost': 'gbdt',
       'feature_fraction': 0.6534414209120409,
       'learning_rate': 0.01,
       'max_bin': 1000,
       'max_depth': 7,
       'min_child_samples': 75,
       'min_child_weight': 0,
       'num_leaves': 15,
       'reg_alpha': 1.0,
       'reg_lambda': 1e-09,
       'scale_pos_weight': 1.0,
       'subsample_for_bin': 500000,
       'num_threads': 5,
       'tree_learner': 'serial',
       'objective': 'binary',
       'verbosity': 1,
       'metric':'auc',
       'seed': 17 }

In [91]:

params = {'boost': 'gbdt',
          'feature_fraction': 0.05,
          'learning_rate': 0.01,
          'max_depth': -1,  
          'metric':'auc',
          'min_data_in_leaf': 50,
          'num_leaves': 32,
          'num_threads': -1,
          'verbosity': 1,
          'objective': 'binary',
          'seed': 17
         }

In [92]:

y = target['radiant_win']

Working with json data¶

In [93]:

with open(os.path.join(PATH, 'train_matches.jsonl')) as fin:
    # read the 18-th line
    for i in range(18):
        line = fin.readline()
    
    # read JSON into a Python object 
    match = json.loads(line)

In [94]:

match.keys()

Out[94]:

dict_keys(['game_time', 'match_id_hash', 'teamfights', 'objectives', 'chat', 'game_mode', 'lobby_type', 'players', 'targets'])

In [95]:

# match['players'][1]

In [96]:

def read_matches(matches_file):
    
    MATCHES_COUNT = {
        'test_matches.jsonl': 10000,
        'train_matches.jsonl': 39675,
    }
    _, filename = os.path.split(matches_file)
    total_matches = MATCHES_COUNT.get(filename)
    
    with open(matches_file) as fin:
        for line in tqdm_notebook(fin, total=total_matches):
            yield json.loads(line)

In [97]:

import collections

MATCH_FEATURES = [
    ('game_time', lambda m: m['game_time']),
    ('game_mode', lambda m: m['game_mode']),
    ('lobby_type', lambda m: m['lobby_type']),
    ('objectives_len', lambda m: len(m['objectives'])),
    ('chat_len', lambda m: len(m['chat'])),
]

PLAYER_FIELDS = [
    'hero_id',
    
    'kills',
    'deaths',
    'assists',
    'denies',
    
    'gold',
    'lh',
    'xp',
    'health',
    'max_health',
    'max_mana',
    'level',

    'x',
    'y',
    
    'stuns',
    'creeps_stacked',
    'camps_stacked',
    'rune_pickups',
    'firstblood_claimed',
    'teamfight_participation',
    'towers_killed',
    'roshans_killed',
    'obs_placed',
    'sen_placed'
]
def sum_features(feature_log):
    sum = 0
    for key in feature_log:
        sum += feature_log[key]
    return sum

def extract_features_csv(match):
    row = [
        ('match_id_hash', match['match_id_hash']),
    ]

    for field, f in MATCH_FEATURES:
        row.append((field, f(match)))
        
    for slot, player in enumerate(match['players']):
        if slot < 5:
            player_name = 'r%d' % (slot + 1)
        else:
            player_name = 'd%d' % (slot - 4)

        row.append( (f'{player_name}_items', list(map(lambda x: x['id'][5:], player['hero_inventory']))))
#         row.append( (f'{player_name}_stash_items', list(map(lambda x: x['id'][5:], player['hero_stash']))))
#         row.append( (f'{player_name}_purchase_items', list(player['purchase'])))

        for field in PLAYER_FIELDS:
            column_name = '%s_%s' % (player_name, field)
            row.append((column_name, player[field]))
        row.append((f'{player_name}_ability_level', len(player['ability_upgrades'])))
        row.append((f'{player_name}_max_hero_hit', player['max_hero_hit']['value']))
        row.append((f'{player_name}_purchase_count', len(player['purchase_log'])))
        row.append((f'{player_name}_count_ability_use', sum(player['ability_uses'].values())))
        row.append((f'{player_name}_damage_dealt', sum(player['damage'].values())))
        row.append((f'{player_name}_damage_received', sum(player['damage_taken'].values())))

# Counting aegis stolen for both teams
        radiant_aegis_stolen = 0
        dire_aegis_stolen = 0
        for objective in match['objectives']:
            if objective['type'] == 'CHAT_MESSAGE_AEGIS_STOLEN':
                if objective['slot'] <=4:
                    radiant_aegis_stolen += 1
                if objective['slot'] >=5:
                    dire_aegis_stolen += 1

        row.append(('radiant_aegis_stolen', radiant_aegis_stolen))
        row.append(('dire_aegis_stolen', dire_aegis_stolen))
        row.append(('diff_aegis', radiant_aegis_stolen - dire_aegis_stolen))
        
# Counting ruined towers for both teams
        radiant_tower_kills = 0
        dire_tower_kills = 0
        for objective in match['objectives']:
            if objective['type'] == 'CHAT_MESSAGE_TOWER_KILL':
                if objective['team'] == 2:
                    radiant_tower_kills += 1
                if objective['team'] == 3:
                    dire_tower_kills += 1

        row.append(('radiant_tower_kills', radiant_tower_kills))
        row.append(('dire_tower_kills', dire_tower_kills))
        row.append(('diff_tower_kills', radiant_tower_kills - dire_tower_kills))
        
# Counting baracks destroed for both teams
        radiant_barracks_kills = 0
        radiant_barracks1_kills = 0
        dire_barracks_kills = 0
        dire_barracks1_kills = 0
        radiant = ['1','2','4','8','16','32']
        dire = ['64','128','256','512','1024','2048']
        for objective in match['objectives']:
            if objective['type'] == 'CHAT_MESSAGE_BARRACKS_KILL':
                if objective['key'] in ['1', '4', '16']:
                    radiant_barracks_kills += 1
                if objective['key'] in ['1', '8', '32']:
                    radiant_barracks1_kills += 1
                if objective['key'] in ['64', '256', '1024']:
                    dire_barracks_kills += 1
                if objective['key'] in ['128', '512', '2048']:
                    dire_barracks1_kills += 1
        row.append(('megacreep', (int(radiant_barracks_kills+radiant_barracks1_kills ==6) - int(dire_barracks_kills+dire_barracks1_kills==6))))
        row.append(('radiant_barracks_kills', radiant_barracks_kills))
        row.append(('dire_barracks_kills', dire_barracks_kills))
        row.append(('diff_barracks_kills', radiant_barracks_kills - dire_barracks_kills))
        row.append(('diff_barracks_kills1', radiant_barracks1_kills - dire_barracks1_kills))
    
    return collections.OrderedDict(row)
    
def extract_targets_csv(match, targets):
    return collections.OrderedDict([('match_id_hash', match['match_id_hash'])] + [
        (field, targets[field])
        for field in ['game_time', 'radiant_win', 'duration', 'time_remaining', 'next_roshan_team']
    ])

In [98]:

# Make reverse data move
def extract_inverse_features_csv(match):
    row = [
        ('match_id_hash', match['match_id_hash'][::-1]),
    ]
    
    for field, f in MATCH_FEATURES:
        row.append((field, f(match)))
        
    for slot, player in enumerate(match['players']):
        if slot < 5:
            player_name = 'd%d' % (slot + 1)
        else:
            player_name = 'r%d' % (slot - 4)
        
        row.append( (f'{player_name}_items', list(map(lambda x: x['id'][5:], player['hero_inventory']))))
#         row.append( (f'{player_name}_stash_items', list(map(lambda x: x['id'][5:], player['hero_stash']))))
#         row.append( (f'{player_name}_purchase_items', list(player['purchase'])))
        for field in PLAYER_FIELDS:
            column_name = '%s_%s' % (player_name, field)
            row.append((column_name, player[field]))
        row.append((f'{player_name}_ability_level', len(player['ability_upgrades'])))
        row.append((f'{player_name}_max_hero_hit', player['max_hero_hit']['value']))
        row.append((f'{player_name}_purchase_count', len(player['purchase_log'])))
        row.append((f'{player_name}_count_ability_use', sum(player['ability_uses'].values())))
        row.append((f'{player_name}_damage_dealt', sum(player['damage'].values())))
        row.append((f'{player_name}_damage_received', sum(player['damage_taken'].values())))

# Counting aegis stolen for both teams
        radiant_aegis_stolen = 0
        dire_aegis_stolen = 0
        for objective in match['objectives']:
            if objective['type'] == 'CHAT_MESSAGE_AEGIS_STOLEN':
                if objective['slot'] >=5:
                    radiant_aegis_stolen += 1
                if objective['slot'] <=4:
                    dire_aegis_stolen += 1

        row.append(('radiant_aegis_stolen', radiant_aegis_stolen))
        row.append(('dire_aegis_stolen', dire_aegis_stolen))
        row.append(('diff_aegis', radiant_aegis_stolen - dire_aegis_stolen))
# Counting ruined towers for both teams
        radiant_tower_kills = 0
        dire_tower_kills = 0
        for objective in match['objectives']:
            if objective['type'] == 'CHAT_MESSAGE_TOWER_KILL':
                if objective['team'] == 3:
                    radiant_tower_kills += 1
                if objective['team'] == 2:
                    dire_tower_kills += 1

        row.append(('radiant_tower_kills', radiant_tower_kills))
        row.append(('dire_tower_kills', dire_tower_kills))
        row.append(('diff_tower_kills', radiant_tower_kills - dire_tower_kills))
        
# Counting baracks destroed for both teams
        radiant_barracks_kills = 0
        radiant_barracks1_kills = 0
        dire_barracks_kills = 0
        dire_barracks1_kills = 0
        radiant = ['1','2','4','8','16','32']
        dire = ['64','128','256','512','1024','2048']
        for objective in match['objectives']:
            if objective['type'] == 'CHAT_MESSAGE_BARRACKS_KILL':
                if objective['key'] in ['1', '4', '16']:
                    dire_barracks_kills += 1
                if objective['key'] in ['1', '8', '32']:
                    dire_barracks1_kills += 1
                if objective['key'] in ['64', '256', '1024']:
                    radiant_barracks_kills += 1
                if objective['key'] in ['128', '512', '2048']:
                    radiant_barracks1_kills += 1
        row.append(('megacreep', (int(radiant_barracks_kills+radiant_barracks1_kills ==6) - int(dire_barracks_kills+dire_barracks1_kills==6))))
        row.append(('radiant_barracks_kills', radiant_barracks_kills))
        row.append(('dire_barracks_kills', dire_barracks_kills))
        row.append(('diff_barracks_kills', radiant_barracks_kills - dire_barracks_kills))
        row.append(('diff_barracks_kills1', radiant_barracks1_kills - dire_barracks1_kills))
       
    return collections.OrderedDict(row)
    
def extract_inverse_targets_csv(match, targets):
    return collections.OrderedDict([('match_id_hash', match['match_id_hash'][::-1])] + [
        (field,  not targets[field])
        for field in ['game_time', 'radiant_win', 'duration', 'time_remaining', 'next_roshan_team']
    ])

In [99]:

%%time

df_features = []
df_targets = []

df_inverse_features = []
df_inverse_targets = []

for match in read_matches(os.path.join(PATH, 'train_matches.jsonl')):
    features = extract_features_csv(match)
    inverse_features = extract_inverse_features_csv(match)
    targets = extract_targets_csv(match, match['targets'])
    inverse_targets = extract_inverse_targets_csv(match, match['targets'])
    
    df_features.append(features)
    df_inverse_features.append(features)
    df_inverse_features.append(inverse_features)
    df_targets.append(targets)
    df_inverse_targets.append(targets)
    df_inverse_targets.append(inverse_targets)

HBox(children=(IntProgress(value=0, max=39675), HTML(value='')))

CPU times: user 1min 6s, sys: 2.8 s, total: 1min 9s
Wall time: 1min 9s

In [100]:

train_df = pd.DataFrame.from_records(df_features).set_index('match_id_hash')
y_train = pd.DataFrame.from_records(df_targets).set_index('match_id_hash')
y_train = y_train['radiant_win'].map({True: 1, False: 0})

In [101]:

df_new_features = pd.DataFrame.from_records(df_inverse_features).set_index('match_id_hash')
y_train_inverse = pd.DataFrame.from_records(df_inverse_targets).set_index('match_id_hash')
y_train_inverse = y_train_inverse['radiant_win'].map({True: 1, False: 0})

In [102]:

df_new_features.shape

Out[102]:

(79350, 326)

In [105]:

# import pickle as pkl
# yt.to_pickle('target_final2_features.pkl')

In [106]:

print(train_df.shape)
print(df_new_features.shape)

(39675, 326)
(79350, 326)

In [107]:

# # Using without double train data
# # %%time
# PATH_TO_DATA = PATH
# df_new_features = []
# df_new_targets = []

# for match in read_matches(os.path.join(PATH_TO_DATA, 'train_matches.jsonl')):
#     match_id_hash = match['match_id_hash']
#     features = extract_features_csv(match)
#     targets = extract_targets_csv(match, match['targets'])
    
#     df_new_features.append(features)
#     df_new_targets.append(targets)

In [108]:

import os
import ujson as json
from tqdm import tqdm_notebook

def read_matches(matches_file):
    
    MATCHES_COUNT = {
        'test_matches.jsonl': 10000,
        'train_matches.jsonl': 39675,
    }
    _, filename = os.path.split(matches_file)
    total_matches = MATCHES_COUNT.get(filename)
    
    with open(matches_file) as fin:
        for line in tqdm_notebook(fin, total=total_matches):
            yield json.loads(line)

def add_new_features(df_features, matches_file):
    
    # Process raw data and add new features
    for match in read_matches(matches_file):
        match_id_hash = match['match_id_hash']
        if (match['game_time'] == 0) and (len(match['players'][0]['times']) > 0):
            df_features.loc[match_id_hash, 'game_time'] = match['players'][0]['times'][-1] + 30
            
        game_time = int(df_features.loc[match_id_hash, 'game_time'])
  
        
        ### Damage features// team damage vs heroes, vs towers and vs creeps
        ### also buybacks for last 5 mins
        ### also observers placed for last 5 mins
        ### also healing
        ### also lifestate
        
        team_damage_vs_other_team = 0
        
        damage_vs_dire_tower = 0
        damage_vs_rad_tower = 0
        
        rad_vs_creeps = 0
        dire_vs_creeps = 0
        
        dire_bb = 0
        rad_bb = 0
        
        dire_obs, rad_obs = 0, 0
        rad_heal ,dire_heal = 0, 0
        rad_alive, rad_dead = 0, 0
        dire_alive, dire_dead = 0, 0
        
        dire_actions, rad_actions = 0, 0
        rad_items, dire_items = 0, 0
        rad_ability, dire_ability = 0, 0 
        
        for i, player in enumerate(match['players']):
            for whom in player['damage'].keys():
                if whom[9:13] == 'hero':
                    team_damage_vs_other_team += player['damage'][whom]
                
                elif (whom[0:22] == 'npc_dota_badguys_tower') and (i < 5):
                    damage_vs_dire_tower += player['damage'][whom]
                elif (whom[0:23] == 'npc_dota_goodguys_tower') and (i >=5):
                    damage_vs_rad_tower += player['damage'][whom]
                elif (whom[0:22] == 'npc_dota_creep_badguys') and (i < 5):
                    rad_vs_creeps += player['damage'][whom]
                elif (whom[0:23] == 'npc_dota_creep_goodguys') and (i >= 5):
                    dire_vs_creeps += player['damage'][whom]
                
            if i == 4:
                df_features.loc[match_id_hash, 'rad_damage_vs_heroes'] = team_damage_vs_other_team
                team_damage_vs_other_team = 0
            
            elif i == 9:
                df_features.loc[match_id_hash, 'dire_damage_vs_heroes'] = team_damage_vs_other_team
        
            
            
            ### buybacks
            if len(player['buyback_log']) > 0:
                  
                if (i < 5) and (game_time < player['buyback_log'][-1]['time'] + 300):
                    rad_bb += 1
                elif (i >= 5) and (game_time < player['buyback_log'][-1]['time'] + 300):
                    dire_bb += 1
                    
            if len(player['obs_log']) > 0:
                for obs in player['obs_log']:  
                    if (i < 5) and (game_time < obs['time'] + 300):
                        rad_obs += 1
                    elif (i >= 5) and (game_time < obs['time'] + 300):
                        dire_obs += 1
            
            if len(player['healing']) > 0:
                if i < 5:
                    rad_heal += sum(list(player['healing'].values()))
                else:
                    dire_heal += sum(list(player['healing'].values()))
                
                
            if i < 5:
                if '0' in player['life_state'].keys():
                    rad_alive += player['life_state']['0']
                if '2' in player['life_state'].keys():
                    rad_dead += player['life_state']['2'] 
            else:
                if '0' in player['life_state'].keys():
                    dire_alive += player['life_state']['0']
                if '2' in player['life_state'].keys():
                    dire_dead += player['life_state']['2']
                
            rad_actions += int(i<5) * sum(list(player['actions'].values()))
            dire_actions += int(i>4) * sum(list(player['actions'].values()))
        
            rad_items += int(i<5) * sum(list(player['item_uses'].values()))
            dire_items += int(i>4) * sum(list(player['item_uses'].values()))
            
            rad_ability += int(i<5) * sum(list(player['ability_uses'].values()))
            dire_ability += int(i>4) * sum(list(player['ability_uses'].values()))
            
        df_features.loc[match_id_hash, 'bb_restr_Diff'] = rad_bb - dire_bb                 ### bb
        
        df_features.loc[match_id_hash, 'obs_on_map_Diff'] = rad_obs - dire_obs             ### obs
        
        df_features.loc[match_id_hash, 'rad_damage_vs_tower'] = damage_vs_dire_tower
        df_features.loc[match_id_hash, 'dire_damage_vs_tower'] = damage_vs_rad_tower
        
        if  (damage_vs_dire_tower + damage_vs_rad_tower) > 0:
            df_features.loc[match_id_hash, 'damage_vs_tower_Percent'] = damage_vs_dire_tower / \
                (damage_vs_dire_tower + damage_vs_rad_tower)                                   ### damage vs tower
        else:
            df_features.loc[match_id_hash, 'damage_vs_tower_Percent']= 0.5
            
        df_features.loc[match_id_hash, 'rad_damage_vs_creeps'] = rad_vs_creeps             
        df_features.loc[match_id_hash, 'dire_damage_vs_creeps'] = dire_vs_creeps
        if  (dire_vs_creeps + rad_vs_creeps)  > 0:            
            df_features.loc[match_id_hash, 'damage_vs_creeps_Percent'] = rad_vs_creeps / \
                (dire_vs_creeps + rad_vs_creeps)                                              ### damage vs creeps
        else:
            df_features.loc[match_id_hash, 'damage_vs_creeps_Percent'] = 0.5 
        
        df_features.loc[match_id_hash, 'healing_Diff'] = (rad_heal - dire_heal)
        if dire_heal + rad_heal > 0:
            df_features.loc[match_id_hash, 'healing_Percent'] = rad_heal / (dire_heal + rad_heal)
                               ### healing
        else:
            df_features.loc[match_id_hash, 'healing_Percent'] = 0.5
        if (rad_alive + dire_alive) != 0:
            df_features.loc[match_id_hash, 'alive_Percent'] = (rad_alive) / (rad_alive + dire_alive)
        else:
            df_features.loc[match_id_hash, 'alive_Percent'] = 0.5
        if (rad_dead + dire_dead) != 0:
            df_features.loc[match_id_hash, 'dead_Percent'] = (rad_dead) / (rad_dead + dire_dead)     ### life_state
        else:
            df_features.loc[match_id_hash, 'dead_Percent'] = 0.5
        
        if (rad_actions + dire_actions) > 0:
            df_features.loc[match_id_hash, 'actions_Percent'] = rad_actions / (rad_actions + dire_actions)
        else:
            df_features.loc[match_id_hash, 'actions_Percent'] = 0.5
            
        df_features.loc[match_id_hash, 'actions_Diff_permin'] = (rad_actions - dire_actions) / (game_time // 60 + 1)
        
        if (rad_items + dire_items) > 0:
            df_features.loc[match_id_hash, 'items_uses_Percent'] = rad_items / (rad_items + dire_items)
        else:
            df_features.loc[match_id_hash, 'items_uses_Percent'] = 0.5
            
        df_features.loc[match_id_hash, 'items_uses_Diff_permin'] = (rad_items - dire_items) / (game_time // 60 + 1)
        
        if (rad_ability + dire_ability) > 0:
            df_features.loc[match_id_hash, 'ability_Percent'] = rad_ability / (rad_ability + dire_ability)
        else:
            df_features.loc[match_id_hash, 'ability_Percent'] = 0.5
            
        df_features.loc[match_id_hash, 'ability_Diff_permin'] = (rad_ability - dire_ability) / (game_time // 60 + 1)

In [109]:

add_new_features(df_new_features, os.path.join(PATH, 'train_matches.jsonl'))

HBox(children=(IntProgress(value=0, max=39675), HTML(value='')))

In [110]:

df_new_features.head()

Out[110]:

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_items	r1_hero_id	r1_kills	r1_deaths	r1_assists	r1_denies	r1_gold	r1_lh	r1_xp	r1_health	r1_max_health	r1_max_mana	r1_level	r1_x	r1_y	r1_stuns	r1_creeps_stacked	r1_camps_stacked	r1_rune_pickups	r1_firstblood_claimed	r1_teamfight_participation	r1_towers_killed	r1_roshans_killed	r1_obs_placed	r1_sen_placed	r1_ability_level	r1_max_hero_hit	r1_purchase_count	r1_count_ability_use	r1_damage_dealt	r1_damage_received	radiant_aegis_stolen	dire_aegis_stolen	diff_aegis	radiant_tower_kills	dire_tower_kills	diff_tower_kills	megacreep	radiant_barracks_kills	dire_barracks_kills	diff_barracks_kills	diff_barracks_kills1	r2_items	r2_hero_id	r2_kills	r2_deaths	r2_assists	r2_denies	r2_gold	r2_lh	r2_xp	r2_health	r2_max_health	r2_max_mana	r2_level	r2_x	r2_y	r2_stuns	r2_creeps_stacked	r2_camps_stacked	r2_rune_pickups	r2_firstblood_claimed	r2_teamfight_participation	r2_towers_killed	r2_roshans_killed	r2_obs_placed	r2_sen_placed	r2_ability_level	r2_max_hero_hit	r2_purchase_count	r2_count_ability_use	r2_damage_dealt	r2_damage_received	r3_items	r3_hero_id	r3_kills	r3_deaths	r3_assists	r3_denies	r3_gold	r3_lh	r3_xp	r3_health	r3_max_health	r3_max_mana	r3_level	r3_x	r3_y	r3_stuns	r3_creeps_stacked	r3_camps_stacked	r3_rune_pickups	r3_firstblood_claimed	r3_teamfight_participation	r3_towers_killed	r3_roshans_killed	r3_obs_placed	r3_sen_placed	r3_ability_level	r3_max_hero_hit	r3_purchase_count	r3_count_ability_use	r3_damage_dealt	r3_damage_received	r4_items	r4_hero_id	r4_kills	r4_deaths	r4_assists	r4_denies	r4_gold	r4_lh	r4_xp	r4_health	r4_max_health	r4_max_mana	r4_level	r4_x	r4_y	r4_stuns	r4_creeps_stacked	r4_camps_stacked	r4_rune_pickups	r4_firstblood_claimed	r4_teamfight_participation	r4_towers_killed	r4_roshans_killed	r4_obs_placed	r4_sen_placed	r4_ability_level	r4_max_hero_hit	r4_purchase_count	r4_count_ability_use	r4_damage_dealt	r4_damage_received	r5_items	r5_hero_id	r5_kills	r5_deaths	r5_assists	r5_denies	r5_gold	r5_lh	r5_xp	r5_health	r5_max_health	r5_max_mana	r5_level	r5_x	r5_y	r5_stuns	r5_creeps_stacked	r5_camps_stacked	r5_rune_pickups	r5_firstblood_claimed	r5_teamfight_participation	r5_towers_killed	r5_roshans_killed	r5_obs_placed	r5_sen_placed	r5_ability_level	r5_max_hero_hit	r5_purchase_count	r5_count_ability_use	r5_damage_dealt	r5_damage_received	d1_items	d1_hero_id	d1_kills	d1_deaths	d1_assists	d1_denies	d1_gold	d1_lh	d1_xp	d1_health	d1_max_health	d1_max_mana	d1_level	d1_x	d1_y	d1_stuns	d1_creeps_stacked	d1_camps_stacked	d1_rune_pickups	d1_firstblood_claimed	d1_teamfight_participation	d1_towers_killed	d1_roshans_killed	d1_obs_placed	d1_sen_placed	d1_ability_level	d1_max_hero_hit	d1_purchase_count	d1_count_ability_use	d1_damage_dealt	d1_damage_received	d2_items	d2_hero_id	d2_kills	d2_deaths	d2_assists	d2_denies	d2_gold	d2_lh	d2_xp	d2_health	d2_max_health	d2_max_mana	d2_level	d2_x	d2_y	d2_stuns	d2_creeps_stacked	d2_camps_stacked	d2_rune_pickups	d2_firstblood_claimed	d2_teamfight_participation	d2_towers_killed	d2_roshans_killed	d2_obs_placed	d2_sen_placed	d2_ability_level	d2_max_hero_hit	d2_purchase_count	d2_count_ability_use	d2_damage_dealt	d2_damage_received	d3_items	d3_hero_id	d3_kills	d3_deaths	d3_assists	d3_denies	d3_gold	d3_lh	d3_xp	d3_health	d3_max_health	d3_max_mana	d3_level	d3_x	d3_y	d3_stuns	d3_creeps_stacked	d3_camps_stacked	d3_rune_pickups	d3_firstblood_claimed	d3_teamfight_participation	d3_towers_killed	d3_roshans_killed	d3_obs_placed	d3_sen_placed	d3_ability_level	d3_max_hero_hit	d3_purchase_count	d3_count_ability_use	d3_damage_dealt	d3_damage_received	d4_items	d4_hero_id	d4_kills	d4_deaths	d4_assists	d4_denies	d4_gold	d4_lh	d4_xp	d4_health	d4_max_health	d4_max_mana	d4_level	d4_x	d4_y	d4_stuns	d4_creeps_stacked	d4_camps_stacked	d4_rune_pickups	d4_firstblood_claimed	d4_teamfight_participation	d4_towers_killed	d4_roshans_killed	d4_obs_placed	d4_sen_placed	d4_ability_level	d4_max_hero_hit	d4_purchase_count	d4_count_ability_use	d4_damage_dealt	d4_damage_received	d5_items	d5_hero_id	d5_kills	d5_deaths	d5_assists	d5_denies	d5_gold	d5_lh	d5_xp	d5_health	d5_max_health	d5_max_mana	d5_level	d5_x	d5_y	d5_stuns	d5_creeps_stacked	d5_camps_stacked	d5_rune_pickups	d5_firstblood_claimed	d5_teamfight_participation	d5_towers_killed	d5_roshans_killed	d5_obs_placed	d5_sen_placed	d5_ability_level	d5_max_hero_hit	d5_purchase_count	d5_count_ability_use	d5_damage_dealt	d5_damage_received	rad_damage_vs_heroes	dire_damage_vs_heroes	bb_restr_Diff	obs_on_map_Diff	rad_damage_vs_tower	dire_damage_vs_tower	damage_vs_tower_Percent	rad_damage_vs_creeps	dire_damage_vs_creeps	damage_vs_creeps_Percent	healing_Diff	healing_Percent	alive_Percent	dead_Percent	actions_Percent	actions_Diff_permin	items_uses_Percent	items_uses_Diff_permin	ability_Percent	ability_Diff_permin
match_id_hash
a400b8f29dece5f4d266f49f1ae2e98a	155	22	7	1	11	[tango, wraith_band, enchanted_mango, clarity,...	11	0	0	0	0	543	7	533	358	600	350.938	2	116	122	0.000	0	0	1	0	0.000	0	0	0	0	0	136	6	3	1287	973	0	0	0	0	0	0	0	0	0	0	0	[]	78	0	0	0	3	399	4	478	636	720	254.938	2	124	126	0.000	0	0	0	0	0.000	0	0	0	0	0	0	6	2	1327	264	[magic_stick, wind_lace]	14	0	1	0	0	304	0	130	700	700	242.938	1	70	156	0.000	0	0	1	0	0.000	0	0	0	0	0	4	5	1	32	1108	[wind_lace, magic_wand, tpscroll]	59	0	0	0	1	389	4	506	399	700	326.938	2	170	86	0.0	0	0	0	0	0.000	0	0	0	0	0	42	9	0	1000	1192	[stout_shield, quelling_blade, branches, faeri...	77	0	0	0	0	402	10	344	422	800	314.938	2	120	100	0.000	0	0	0	0	0.000	0	0	0	0	0	0	7	1	5090	705	[tango, flask, enchanted_mango, stout_shield, ...	12	0	0	1	13	982	12	780	650	720	386.938	3	82	170	0.000	0	0	1	0	1.000	0	0	0	0	0	76	7	1	3521	347	[orb_of_venom, tango, flask, magic_wand, tpscr...	21	0	0	0	6	788	9	706	640	640	422.938	3	174	90	0.000	0	0	2	0	0.000	0	0	0	0	0	50	8	0	1977	190	[orb_of_venom, stout_shield, boots, tango, tps...	60	0	0	0	1	531	0	307	720	720	242.938	2	180	84	0.300	0	0	2	0	0.000	0	0	0	0	0	68	5	3	747	321	[tango, orb_of_venom, boots, branches, branche...	84	1	0	0	0	796	0	421	760	760	326.938	2	90	150	0.0	0	0	2	1	1.000	0	0	1	0	0	62	9	3	868	16	[null_talisman, faerie_fire, tango_single, tps...	34	0	0	0	0	851	11	870	593	680	566.938	3	128	128	0.000	0	0	0	0	0.000	0	0	0	0	0	164	6	4	2332	681	948.0	2949.0	0.0	-1.0	0.0	0.0	0.500000	2180.0	4090.0	0.347687	-289.0	0.003436	0.498382	1.000000	0.591159	409.666667	0.625000	2.666667	0.388889	-1.333333
a89e2ea1f94f662d4f5eced92f8b004a	155	22	7	1	11	[tango, flask, enchanted_mango, stout_shield, ...	12	0	0	1	13	982	12	780	650	720	386.938	3	82	170	0.000	0	0	1	0	1.000	0	0	0	0	0	76	7	1	3521	347	0	0	0	0	0	0	0	0	0	0	0	[orb_of_venom, tango, flask, magic_wand, tpscr...	21	0	0	0	6	788	9	706	640	640	422.938	3	174	90	0.000	0	0	2	0	0.000	0	0	0	0	0	50	8	0	1977	190	[orb_of_venom, stout_shield, boots, tango, tps...	60	0	0	0	1	531	0	307	720	720	242.938	2	180	84	0.300	0	0	2	0	0.000	0	0	0	0	0	68	5	3	747	321	[tango, orb_of_venom, boots, branches, branche...	84	1	0	0	0	796	0	421	760	760	326.938	2	90	150	0.0	0	0	2	1	1.000	0	0	1	0	0	62	9	3	868	16	[null_talisman, faerie_fire, tango_single, tps...	34	0	0	0	0	851	11	870	593	680	566.938	3	128	128	0.000	0	0	0	0	0.000	0	0	0	0	0	164	6	4	2332	681	[tango, wraith_band, enchanted_mango, clarity,...	11	0	0	0	0	543	7	533	358	600	350.938	2	116	122	0.000	0	0	1	0	0.000	0	0	0	0	0	136	6	3	1287	973	[]	78	0	0	0	3	399	4	478	636	720	254.938	2	124	126	0.000	0	0	0	0	0.000	0	0	0	0	0	0	6	2	1327	264	[magic_stick, wind_lace]	14	0	1	0	0	304	0	130	700	700	242.938	1	70	156	0.000	0	0	1	0	0.000	0	0	0	0	0	4	5	1	32	1108	[wind_lace, magic_wand, tpscroll]	59	0	0	0	1	389	4	506	399	700	326.938	2	170	86	0.0	0	0	0	0	0.000	0	0	0	0	0	42	9	0	1000	1192	[stout_shield, quelling_blade, branches, faeri...	77	0	0	0	0	402	10	344	422	800	314.938	2	120	100	0.000	0	0	0	0	0.000	0	0	0	0	0	0	7	1	5090	705	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	[phase_boots, magic_wand, blade_of_alacrity, t...	15	7	2	0	7	5257	52	3937	1160	1160	566.938	8	76	78	0.000	0	0	0	0	0.438	0	0	0	0	6	216	28	9	20882	4857	0	0	0	2	0	2	0	0	0	0	0	[stout_shield, soul_ring, tranquil_boots, chai...	96	3	1	2	3	3394	19	3897	1352	1380	386.938	8	78	166	8.398	0	0	4	0	0.312	0	0	0	0	6	247	19	21	14983	7114	[magic_wand, dust, boots, ward_dispenser, tpsc...	27	1	1	4	2	2212	4	2561	710	860	530.938	6	156	146	11.965	2	1	4	0	0.312	0	0	3	1	4	153	27	11	5941	1758	[magic_wand, pers, blight_stone, tpscroll, rin...	63	4	0	3	12	4206	38	4459	420	880	482.938	9	154	148	0.0	0	0	3	0	0.438	0	0	1	2	7	201	21	35	21394	5410	[boots, stout_shield, tpscroll]	89	1	0	5	4	3103	14	2712	856	900	446.938	6	150	148	21.697	0	0	2	0	0.375	1	0	0	0	5	152	9	18	6565	2064	[magic_wand, power_treads]	58	1	2	0	4	2823	24	3281	700	700	686.938	7	88	170	3.166	1	1	3	0	0.250	0	0	1	0	4	114	15	11	13575	3096	[tranquil_boots, smoke_of_deceit, tpscroll, ma...	14	1	6	0	1	2466	17	2360	758	1040	326.938	6	156	98	0.067	0	0	1	1	0.250	0	0	4	2	4	182	19	45	9643	7316	[power_treads, ward_sentry, branches, ring_of_...	1	1	3	1	7	3624	29	3418	485	800	350.938	7	124	144	0.300	2	1	4	0	0.500	0	0	0	0	5	226	20	22	13855	6000	[boots, magic_wand, wraith_band]	56	0	3	2	3	2808	18	2730	567	1160	410.938	6	124	142	0.0	0	0	6	0	0.500	0	0	0	0	5	78	23	17	7437	5893	[]	92	0	2	0	1	1423	8	1136	800	800	446.938	4	180	176	0.000	0	0	0	0	0.000	0	0	0	0	4	164	11	7	2308	2154	17463.0	13063.0	-1.0	-2.0	1970.0	12.0	0.993946	41940.0	12526.0	0.770022	-99.0	0.486901	0.509275	0.290393	0.724752	1016.818182	0.605882	3.272727	0.479592	-0.727273
d441caf69ec9c97fa2a47ec054c75c9b	658	4	0	3	10	[magic_wand, power_treads]	58	1	2	0	4	2823	24	3281	700	700	686.938	7	88	170	3.166	1	1	3	0	0.250	0	0	1	0	4	114	15	11	13575	3096	0	0	0	0	2	-2	0	0	0	0	0	[tranquil_boots, smoke_of_deceit, tpscroll, ma...	14	1	6	0	1	2466	17	2360	758	1040	326.938	6	156	98	0.067	0	0	1	1	0.250	0	0	4	2	4	182	19	45	9643	7316	[power_treads, ward_sentry, branches, ring_of_...	1	1	3	1	7	3624	29	3418	485	800	350.938	7	124	144	0.300	2	1	4	0	0.500	0	0	0	0	5	226	20	22	13855	6000	[boots, magic_wand, wraith_band]	56	0	3	2	3	2808	18	2730	567	1160	410.938	6	124	142	0.0	0	0	6	0	0.500	0	0	0	0	5	78	23	17	7437	5893	[]	92	0	2	0	1	1423	8	1136	800	800	446.938	4	180	176	0.000	0	0	0	0	0.000	0	0	0	0	4	164	11	7	2308	2154	[phase_boots, magic_wand, blade_of_alacrity, t...	15	7	2	0	7	5257	52	3937	1160	1160	566.938	8	76	78	0.000	0	0	0	0	0.438	0	0	0	0	6	216	28	9	20882	4857	[stout_shield, soul_ring, tranquil_boots, chai...	96	3	1	2	3	3394	19	3897	1352	1380	386.938	8	78	166	8.398	0	0	4	0	0.312	0	0	0	0	6	247	19	21	14983	7114	[magic_wand, dust, boots, ward_dispenser, tpsc...	27	1	1	4	2	2212	4	2561	710	860	530.938	6	156	146	11.965	2	1	4	0	0.312	0	0	3	1	4	153	27	11	5941	1758	[magic_wand, pers, blight_stone, tpscroll, rin...	63	4	0	3	12	4206	38	4459	420	880	482.938	9	154	148	0.0	0	0	3	0	0.438	0	0	1	2	7	201	21	35	21394	5410	[boots, stout_shield, tpscroll]	89	1	0	5	4	3103	14	2712	856	900	446.938	6	150	148	21.697	0	0	2	0	0.375	1	0	0	0	5	152	9	18	6565	2064	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
6db558535151ea18ca70a6892197db41	21	23	0	0	0	[null_talisman, branches, branches, tpscroll]	101	0	0	0	0	176	0	0	680	680	506.938	1	118	118	0.000	0	0	0	0	0.000	0	0	0	0	0	0	4	0	0	0	0	0	0	0	0	0	0	0	0	0	0	[tango, flask, stout_shield, clarity, enchante...	51	0	0	0	0	176	0	0	720	720	278.938	1	156	104	0.000	0	0	0	0	0.000	0	0	0	0	0	0	7	0	0	0	[stout_shield, blight_stone, tpscroll]	44	0	0	0	0	176	0	0	568	600	254.938	1	78	144	0.000	0	0	1	0	0.000	0	0	0	0	0	70	3	1	70	68	[stout_shield, quelling_blade, tango, tpscroll]	49	0	0	0	0	176	0	0	580	580	254.938	1	150	78	0.0	0	0	1	0	0.000	0	0	0	0	0	0	4	0	0	0	[tango, faerie_fire, tpscroll]	53	0	0	0	0	176	0	0	580	580	374.938	1	78	142	0.000	0	0	1	0	0.000	0	0	0	0	0	0	3	0	0	0	[tpscroll]	18	0	0	0	0	96	0	0	660	660	266.938	1	180	178	0.000	0	0	0	0	0.000	0	0	0	0	0	0	1	0	0	0	[tango, quelling_blade, stout_shield, branches...	67	0	0	0	0	96	0	0	586	620	278.938	1	100	174	0.000	0	0	0	0	0.000	0	0	0	0	0	68	5	1	68	70	[tango, wraith_band, tpscroll]	47	0	0	0	0	96	0	0	660	660	290.938	1	178	112	0.000	0	0	1	0	0.000	0	0	0	0	0	0	3	0	0	0	[flask, tango, stout_shield, circlet, tpscroll]	40	0	0	0	0	96	0	0	600	600	302.938	1	176	110	0.0	0	0	0	0	0.000	0	0	0	0	0	0	5	0	0	0	[null_talisman, clarity, ward_observer]	17	0	0	0	0	96	0	0	640	640	446.938	1	162	162	0.000	0	0	0	0	0.000	0	0	0	0	0	0	4	0	0	0	70.0	68.0	0.0	0.0	0.0	0.0	0.500000	0.0	0.0	0.500000	0.0	0.500000	0.500000	0.500000	0.500000	0.000000	0.000000	-1.000000	0.500000	0.000000

In [111]:

# df_new_features = pd.DataFrame.from_records(df_new_features).set_index('match_id_hash')
# df_new_targets = pd.DataFrame.from_records(df_new_targets).set_index('match_id_hash')

In [112]:

# df_new_features.head()

In [113]:

df_new_features = df_new_features.fillna(0)

In [114]:

# df_new_features[df_new_features['game_time'] == 0]

In [115]:

# df_new_features['dire_aegis_stolen'].value_counts()

In [116]:

test_new_features = []
for match in read_matches(os.path.join(PATH, 'test_matches.jsonl')):
    match_id_hash = match['match_id_hash']
    features = extract_features_csv(match)
    
    test_new_features.append(features)
test_new_features = pd.DataFrame.from_records(test_new_features).set_index('match_id_hash')

HBox(children=(IntProgress(value=0, max=10000), HTML(value='')))

In [117]:

add_new_features(test_new_features, os.path.join(PATH, 'test_matches.jsonl'))

HBox(children=(IntProgress(value=0, max=10000), HTML(value='')))

In [118]:

test_new_features = test_new_features.fillna(0)

In [119]:

df_new_features.shape, test_new_features.shape

Out[119]:

((79350, 346), (10000, 346))

In [120]:

# Adding items
def add_items_dummies(df_new_features, test_new_features):
    
    full_df = pd.concat([df_new_features, test_new_features], sort=False)
    train_size = df_new_features.shape[0]

    for team in 'r', 'd':
        players = [f'{team}{i}' for i in range(1, 6)]
        item_columns = [f'{player}_items' for player in players]

        d = pd.get_dummies(full_df[item_columns[0]].apply(pd.Series).stack()).sum(level=0, axis=0)
        dindexes = d.index.values

        for c in item_columns[1:]:
            d = d.add(pd.get_dummies(full_df[c].apply(pd.Series).stack()).sum(level=0, axis=0), fill_value=0)
            d = d.ix[dindexes]
 
        full_df = pd.concat([full_df, d.add_prefix(f'{team}_item_')], axis=1, sort=False)
        full_df.drop(columns=item_columns, inplace=True)

    df_new_features = full_df.iloc[:train_size, :]
    test_new_features = full_df.iloc[train_size:, :]

    return df_new_features, test_new_features

In [121]:

# Drop consumble items
def drop_consumble_items(df_new_features, test_new_features):
    
    full_df = pd.concat([df_new_features, test_new_features], sort=False)
    train_size = df_new_features.shape[0]

    for team in 'r', 'd':
        consumble_columns = ['tango', 'tpscroll', 
                             'bottle', 'flask',
                            'enchanted_mango', 'clarity',
                            'faerie_fire', 'ward_observer',
                            'ward_sentry']
        
        starts_with = f'{team}_item_'
        consumble_columns = [starts_with + column for column in consumble_columns]
        full_df.drop(columns=consumble_columns, inplace=True)

    df_new_features = full_df.iloc[:train_size, :]
    test_new_features = full_df.iloc[train_size:, :]

    return df_new_features, test_new_features

In [122]:

%%time
df_new_features, test_new_features = add_items_dummies(df_new_features, test_new_features)
df_new_features, test_new_features = drop_consumble_items(df_new_features, test_new_features)

CPU times: user 4min 13s, sys: 23.5 s, total: 4min 36s
Wall time: 4min 32s

In [123]:

df_new_features.head()

Out[123]:

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_kills	r1_deaths	r1_assists	r1_denies	r1_gold	r1_lh	r1_xp	r1_health	r1_max_health	r1_max_mana	r1_level	r1_x	r1_y	r1_stuns	r1_creeps_stacked	r1_camps_stacked	r1_rune_pickups	r1_teamfight_participation	r1_obs_placed	r1_ability_level	r1_max_hero_hit	r1_purchase_count	r1_count_ability_use	r1_damage_dealt	r1_damage_received	radiant_tower_kills	dire_tower_kills	diff_tower_kills	r2_hero_id	r2_kills	r2_deaths	r2_assists	r2_denies	r2_gold	r2_lh	r2_xp	r2_health	r2_max_health	r2_max_mana	r2_level	r2_x	r2_y	r2_stuns	r2_rune_pickups	r2_firstblood_claimed	r2_teamfight_participation	r2_obs_placed	r2_sen_placed	r2_ability_level	r2_max_hero_hit	r2_purchase_count	r2_count_ability_use	r2_damage_dealt	r2_damage_received	r3_hero_id	r3_kills	r3_deaths	r3_assists	r3_denies	r3_gold	r3_lh	r3_xp	r3_health	r3_max_health	r3_max_mana	r3_level	r3_x	r3_y	r3_stuns	r3_creeps_stacked	r3_camps_stacked	r3_rune_pickups	r3_teamfight_participation	r3_obs_placed	r3_sen_placed	r3_ability_level	r3_max_hero_hit	r3_purchase_count	r3_count_ability_use	r3_damage_dealt	r3_damage_received	r4_hero_id	r4_kills	r4_deaths	r4_assists	r4_denies	r4_gold	r4_lh	r4_xp	r4_health	r4_max_health	r4_max_mana	r4_level	r4_x	r4_y	r4_rune_pickups	r4_firstblood_claimed	r4_teamfight_participation	r4_obs_placed	r4_sen_placed	r4_ability_level	r4_max_hero_hit	r4_purchase_count	r4_count_ability_use	r4_damage_dealt	r4_damage_received	r5_hero_id	r5_kills	r5_deaths	r5_assists	r5_denies	r5_gold	r5_lh	r5_xp	r5_health	r5_max_health	r5_max_mana	r5_level	r5_x	r5_y	r5_stuns	r5_rune_pickups	r5_teamfight_participation	r5_towers_killed	r5_ability_level	r5_max_hero_hit	r5_purchase_count	r5_count_ability_use	r5_damage_dealt	r5_damage_received	d1_hero_id	d1_kills	d1_deaths	d1_assists	d1_denies	d1_gold	d1_lh	d1_xp	d1_health	d1_max_health	d1_max_mana	d1_level	d1_x	d1_y	d1_stuns	d1_creeps_stacked	d1_camps_stacked	d1_rune_pickups	d1_teamfight_participation	d1_obs_placed	d1_ability_level	d1_max_hero_hit	d1_purchase_count	d1_count_ability_use	d1_damage_dealt	d1_damage_received	d2_hero_id	d2_kills	d2_deaths	d2_assists	d2_denies	d2_gold	d2_lh	d2_xp	d2_health	d2_max_health	d2_max_mana	d2_level	d2_x	d2_y	d2_stuns	d2_rune_pickups	d2_firstblood_claimed	d2_teamfight_participation	d2_obs_placed	d2_sen_placed	d2_ability_level	d2_max_hero_hit	d2_purchase_count	d2_count_ability_use	d2_damage_dealt	d2_damage_received	d3_hero_id	d3_kills	d3_deaths	d3_assists	d3_denies	d3_gold	d3_lh	d3_xp	d3_health	d3_max_health	d3_max_mana	d3_level	d3_x	d3_y	d3_stuns	d3_creeps_stacked	d3_camps_stacked	d3_rune_pickups	d3_teamfight_participation	d3_obs_placed	d3_sen_placed	d3_ability_level	d3_max_hero_hit	d3_purchase_count	d3_count_ability_use	d3_damage_dealt	d3_damage_received	d4_hero_id	d4_kills	d4_deaths	d4_assists	d4_denies	d4_gold	d4_lh	d4_xp	d4_health	d4_max_health	d4_max_mana	d4_level	d4_x	d4_y	d4_rune_pickups	d4_firstblood_claimed	d4_teamfight_participation	d4_obs_placed	d4_sen_placed	d4_ability_level	d4_max_hero_hit	d4_purchase_count	d4_count_ability_use	d4_damage_dealt	d4_damage_received	d5_hero_id	d5_kills	d5_deaths	d5_assists	d5_denies	d5_gold	d5_lh	d5_xp	d5_health	d5_max_health	d5_max_mana	d5_level	d5_x	d5_y	d5_stuns	d5_rune_pickups	d5_teamfight_participation	d5_towers_killed	d5_ability_level	d5_max_hero_hit	d5_purchase_count	d5_count_ability_use	d5_damage_dealt	d5_damage_received	rad_damage_vs_heroes	dire_damage_vs_heroes	bb_restr_Diff	obs_on_map_Diff	rad_damage_vs_tower	dire_damage_vs_tower	damage_vs_tower_Percent	rad_damage_vs_creeps	dire_damage_vs_creeps	damage_vs_creeps_Percent	healing_Diff	healing_Percent	alive_Percent	dead_Percent	actions_Percent	actions_Diff_permin	items_uses_Percent	items_uses_Diff_permin	ability_Percent	ability_Diff_permin	r_item_ancient_janggo	r_item_blade_of_alacrity	r_item_blight_stone	r_item_boots	r_item_branches	r_item_chainmail	r_item_dust	r_item_magic_stick	r_item_magic_wand	r_item_null_talisman	r_item_orb_of_venom	r_item_pers	r_item_phase_boots	r_item_power_treads	r_item_quelling_blade	r_item_recipe_bloodthorn	r_item_recipe_crimson_guard	r_item_recipe_mekansm	r_item_recipe_pipe	r_item_ring_of_aquila	r_item_ring_of_health	r_item_smoke_of_deceit	r_item_soul_ring	r_item_stout_shield	r_item_tango_single	r_item_tranquil_boots	r_item_ward_dispenser	r_item_wind_lace	r_item_wraith_band	d_item_ancient_janggo	d_item_blade_of_alacrity	d_item_blight_stone	d_item_boots	d_item_branches	d_item_chainmail	d_item_circlet	d_item_dust	d_item_magic_stick	d_item_magic_wand	d_item_null_talisman	d_item_orb_of_venom	d_item_pers	d_item_phase_boots	d_item_power_treads	d_item_quelling_blade	d_item_recipe_bloodthorn	d_item_recipe_crimson_guard	d_item_recipe_pipe	d_item_ring_of_aquila	d_item_ring_of_health	d_item_smoke_of_deceit	d_item_soul_ring	d_item_stout_shield	d_item_tango_single	d_item_tranquil_boots	d_item_ward_dispenser	d_item_wind_lace	d_item_wraith_band
a400b8f29dece5f4d266f49f1ae2e98a	155	22	7	1	11	11	0	0	0	0	543	7	533	358	600	350.938	2	116	122	0.000	0	0	1	0.000	0	0	136	6	3	1287	973	0	0	0	78	0	0	0	3	399	4	478	636	720	254.938	2	124	126	0.000	0	0	0.000	0	0	0	0	6	2	1327	264	14	0	1	0	0	304	0	130	700	700	242.938	1	70	156	0.000	0	0	1	0.000	0	0	0	4	5	1	32	1108	59	0	0	0	1	389	4	506	399	700	326.938	2	170	86	0	0	0.000	0	0	0	42	9	0	1000	1192	77	0	0	0	0	402	10	344	422	800	314.938	2	120	100	0.000	0	0.000	0	0	0	7	1	5090	705	12	0	0	1	13	982	12	780	650	720	386.938	3	82	170	0.000	0	0	1	1.000	0	0	76	7	1	3521	347	21	0	0	0	6	788	9	706	640	640	422.938	3	174	90	0.000	2	0	0.000	0	0	0	50	8	0	1977	190	60	0	0	0	1	531	0	307	720	720	242.938	2	180	84	0.300	0	0	2	0.000	0	0	0	68	5	3	747	321	84	1	0	0	0	796	0	421	760	760	326.938	2	90	150	2	1	1.000	1	0	0	62	9	3	868	16	34	0	0	0	0	851	11	870	593	680	566.938	3	128	128	0.000	0	0.000	0	0	164	6	4	2332	681	948.0	2949.0	0.0	-1.0	0.0	0.0	0.500000	2180.0	4090.0	0.347687	-289.0	0.003436	0.498382	1.000000	0.591159	409.666667	0.625000	2.666667	0.388889	-1.333333	0.0	0.0	0.0	0.0	2.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	1.0	NaN	0.0	0.0	NaN	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	2.0	1.0	0.0	0.0	0.0	2.0	3.0	0.0	0.0	0.0	0.0	1.0	1.0	3.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0	1.0	0.0	0.0	0.0	0.0
a89e2ea1f94f662d4f5eced92f8b004a	155	22	7	1	11	12	0	0	1	13	982	12	780	650	720	386.938	3	82	170	0.000	0	0	1	1.000	0	0	76	7	1	3521	347	0	0	0	21	0	0	0	6	788	9	706	640	640	422.938	3	174	90	0.000	2	0	0.000	0	0	0	50	8	0	1977	190	60	0	0	0	1	531	0	307	720	720	242.938	2	180	84	0.300	0	0	2	0.000	0	0	0	68	5	3	747	321	84	1	0	0	0	796	0	421	760	760	326.938	2	90	150	2	1	1.000	1	0	0	62	9	3	868	16	34	0	0	0	0	851	11	870	593	680	566.938	3	128	128	0.000	0	0.000	0	0	164	6	4	2332	681	11	0	0	0	0	543	7	533	358	600	350.938	2	116	122	0.000	0	0	1	0.000	0	0	136	6	3	1287	973	78	0	0	0	3	399	4	478	636	720	254.938	2	124	126	0.000	0	0	0.000	0	0	0	0	6	2	1327	264	14	0	1	0	0	304	0	130	700	700	242.938	1	70	156	0.000	0	0	1	0.000	0	0	0	4	5	1	32	1108	59	0	0	0	1	389	4	506	399	700	326.938	2	170	86	0	0	0.000	0	0	0	42	9	0	1000	1192	77	0	0	0	0	402	10	344	422	800	314.938	2	120	100	0.000	0	0.000	0	0	0	7	1	5090	705	0.0	0.0	0.0	0.0	0.0	0.0	0.000000	0.0	0.0	0.000000	0.0	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.0	0.0	0.0	2.0	3.0	0.0	0.0	0.0	1.0	1.0	3.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0	0.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	1.0	NaN	0.0	NaN	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	2.0	1.0
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	15	7	2	0	7	5257	52	3937	1160	1160	566.938	8	76	78	0.000	0	0	0	0.438	0	6	216	28	9	20882	4857	2	0	2	96	3	1	2	3	3394	19	3897	1352	1380	386.938	8	78	166	8.398	4	0	0.312	0	0	6	247	19	21	14983	7114	27	1	1	4	2	2212	4	2561	710	860	530.938	6	156	146	11.965	2	1	4	0.312	3	1	4	153	27	11	5941	1758	63	4	0	3	12	4206	38	4459	420	880	482.938	9	154	148	3	0	0.438	1	2	7	201	21	35	21394	5410	89	1	0	5	4	3103	14	2712	856	900	446.938	6	150	148	21.697	2	0.375	1	5	152	9	18	6565	2064	58	1	2	0	4	2823	24	3281	700	700	686.938	7	88	170	3.166	1	1	3	0.250	1	4	114	15	11	13575	3096	14	1	6	0	1	2466	17	2360	758	1040	326.938	6	156	98	0.067	1	1	0.250	4	2	4	182	19	45	9643	7316	1	1	3	1	7	3624	29	3418	485	800	350.938	7	124	144	0.300	2	1	4	0.500	0	0	5	226	20	22	13855	6000	56	0	3	2	3	2808	18	2730	567	1160	410.938	6	124	142	6	0	0.500	0	0	5	78	23	17	7437	5893	92	0	2	0	1	1423	8	1136	800	800	446.938	4	180	176	0.000	0	0.000	0	4	164	11	7	2308	2154	17463.0	13063.0	-1.0	-2.0	1970.0	12.0	0.993946	41940.0	12526.0	0.770022	-99.0	0.486901	0.509275	0.290393	0.724752	1016.818182	0.605882	3.272727	0.479592	-0.727273	1.0	1.0	1.0	2.0	0.0	1.0	2.0	0.0	3.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	1.0	2.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	1.0	2.0	0.0	0.0	0.0	0.0	2.0	1.0	0.0	NaN	0.0	0.0	1.0	1.0	0.0	1.0	0.0	1.0	0.0	0.0	1.0
d441caf69ec9c97fa2a47ec054c75c9b	658	4	0	3	10	58	1	2	0	4	2823	24	3281	700	700	686.938	7	88	170	3.166	1	1	3	0.250	1	4	114	15	11	13575	3096	0	2	-2	14	1	6	0	1	2466	17	2360	758	1040	326.938	6	156	98	0.067	1	1	0.250	4	2	4	182	19	45	9643	7316	1	1	3	1	7	3624	29	3418	485	800	350.938	7	124	144	0.300	2	1	4	0.500	0	0	5	226	20	22	13855	6000	56	0	3	2	3	2808	18	2730	567	1160	410.938	6	124	142	6	0	0.500	0	0	5	78	23	17	7437	5893	92	0	2	0	1	1423	8	1136	800	800	446.938	4	180	176	0.000	0	0.000	0	4	164	11	7	2308	2154	15	7	2	0	7	5257	52	3937	1160	1160	566.938	8	76	78	0.000	0	0	0	0.438	0	6	216	28	9	20882	4857	96	3	1	2	3	3394	19	3897	1352	1380	386.938	8	78	166	8.398	4	0	0.312	0	0	6	247	19	21	14983	7114	27	1	1	4	2	2212	4	2561	710	860	530.938	6	156	146	11.965	2	1	4	0.312	3	1	4	153	27	11	5941	1758	63	4	0	3	12	4206	38	4459	420	880	482.938	9	154	148	3	0	0.438	1	2	7	201	21	35	21394	5410	89	1	0	5	4	3103	14	2712	856	900	446.938	6	150	148	21.697	2	0.375	1	5	152	9	18	6565	2064	0.0	0.0	0.0	0.0	0.0	0.0	0.000000	0.0	0.0	0.000000	0.0	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.0	0.0	0.0	1.0	1.0	0.0	0.0	1.0	2.0	0.0	0.0	0.0	0.0	2.0	1.0	0.0	NaN	NaN	0.0	0.0	1.0	1.0	0.0	1.0	0.0	1.0	0.0	0.0	1.0	1.0	1.0	1.0	2.0	0.0	1.0	0.0	2.0	0.0	3.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	1.0	2.0	0.0	1.0	1.0	0.0	0.0
6db558535151ea18ca70a6892197db41	21	23	0	0	0	101	0	0	0	0	176	0	0	680	680	506.938	1	118	118	0.000	0	0	0	0.000	0	0	0	4	0	0	0	0	0	0	51	0	0	0	0	176	0	0	720	720	278.938	1	156	104	0.000	0	0	0.000	0	0	0	0	7	0	0	0	44	0	0	0	0	176	0	0	568	600	254.938	1	78	144	0.000	0	0	1	0.000	0	0	0	70	3	1	70	68	49	0	0	0	0	176	0	0	580	580	254.938	1	150	78	1	0	0.000	0	0	0	0	4	0	0	0	53	0	0	0	0	176	0	0	580	580	374.938	1	78	142	0.000	1	0.000	0	0	0	3	0	0	0	18	0	0	0	0	96	0	0	660	660	266.938	1	180	178	0.000	0	0	0	0.000	0	0	0	1	0	0	0	67	0	0	0	0	96	0	0	586	620	278.938	1	100	174	0.000	0	0	0.000	0	0	0	68	5	1	68	70	47	0	0	0	0	96	0	0	660	660	290.938	1	178	112	0.000	0	0	1	0.000	0	0	0	0	3	0	0	0	40	0	0	0	0	96	0	0	600	600	302.938	1	176	110	0	0	0.000	0	0	0	0	5	0	0	0	17	0	0	0	0	96	0	0	640	640	446.938	1	162	162	0.000	0	0.000	0	0	0	4	0	0	0	70.0	68.0	0.0	0.0	0.0	0.0	0.500000	0.0	0.0	0.500000	0.0	0.500000	0.500000	0.500000	0.500000	0.000000	0.000000	-1.000000	0.500000	0.000000	0.0	0.0	1.0	0.0	2.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	3.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	1.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0	0.0	0.0	0.0	0.0	1.0

In [124]:

for c in ['kills', 'deaths', 'assists', 'denies', 'gold', 'lh', 'xp', 'health', 'max_health', 'max_mana', 'level', 'x', 'y', 'stuns', 'creeps_stacked', 'camps_stacked', 'rune_pickups',
          'firstblood_claimed', 'teamfight_participation', 'towers_killed', 'roshans_killed', 'obs_placed', 'sen_placed', 'ability_level', 'max_hero_hit', 'purchase_count',
          'count_ability_use', 'damage_dealt', 'damage_received']:
    r_columns = [f'r{i}_{c}' for i in range(1, 6)]
    d_columns = [f'd{i}_{c}' for i in range(1, 6)]
    
    df_new_features['r_total_' + c] = df_new_features[r_columns].sum(1)
    df_new_features['d_total_' + c] = df_new_features[d_columns].sum(1)
    df_new_features['total_' + c + '_ratio'] = df_new_features['r_total_' + c] / df_new_features['d_total_' + c]
    
    test_new_features['r_total_' + c] = test_new_features[r_columns].sum(1)
    test_new_features['d_total_' + c] = test_new_features[d_columns].sum(1)
    test_new_features['total_' + c + '_ratio'] = test_new_features['r_total_' + c] / test_new_features['d_total_' + c]
    
    df_new_features['r_std_' + c] = df_new_features[r_columns].std(1)
    df_new_features['d_std_' + c] = df_new_features[d_columns].std(1)
    df_new_features['std_' + c + '_ratio'] = df_new_features['r_std_' + c] / df_new_features['d_std_' + c]
    
    test_new_features['r_std_' + c] = test_new_features[r_columns].std(1)
    test_new_features['d_std_' + c] = test_new_features[d_columns].std(1)
    test_new_features['std_' + c + '_ratio'] = test_new_features['r_std_' + c] / test_new_features['d_std_' + c]

Now let's try building a new model!

In [125]:

# Create hero ID
hero_columns = [c for c in df_new_features.columns if '_hero_' in c]
# train = train[hero_columns]
hero_columns2 = [c for c in test_new_features.columns if '_hero_' in c]
# test = test[hero_columns]

for team in 'r', 'd':
    players = [f'{team}{i}' for i in range(1, 6)]
    hero_columns = [f'{player}_hero_id' for player in players]
    d = pd.get_dummies(df_new_features[hero_columns[0]])
    for c in hero_columns[1:]:
        d += pd.get_dummies(df_new_features[c])
    df_new_features = pd.concat([df_new_features, d.add_prefix(f'{team}_hero_')], axis=1)

    
for team in 'r', 'd':
    players = [f'{team}{i}' for i in range(1, 6)]
    hero_columns2 = [f'{player}_hero_id' for player in players]
    d = pd.get_dummies(test_new_features[hero_columns2[0]])
    for c in hero_columns2[1:]:
        d += pd.get_dummies(test_new_features[c])
    test_new_features = pd.concat([test_new_features, d.add_prefix(f'{team}_hero_')], axis=1)

In [126]:

# Adding coordinate for teams
def make_coordinate_features(df):
    for team in 'r', 'd':
        players = [f'{team}{i}' for i in range(1, 6)]
        for player in players:
            df[f'{player}_distance'] = np.sqrt(df[f'{player}_x']**2 + df[f'{player}_y']**2)
            df.drop(columns=[f'{player}_x', f'{player}_y'], inplace=True)
    return df

In [127]:

df_new_features = make_coordinate_features(df_new_features)
test_new_features = make_coordinate_features(test_new_features)

In [128]:

# Combine numeric and drops duplicates
def combine_numeric_features (df, feature_suffixes):
    for feat_suff in feature_suffixes:
        for team in 'r', 'd':
            players = [f'{team}{i}' for i in range(1, 6)] # r1, r2...
            player_col_names = [f'{player}_{feat_suff}' for player in players] # e.g. r1_gold, r2_gold
            
            df[f'{team}_{feat_suff}_max'] = df[player_col_names].max(axis=1) # e.g. r_gold_max
            df[f'{team}_{feat_suff}_min'] = df[player_col_names].min(axis=1) # e.g. r_gold_min
            
            df.drop(columns=player_col_names, inplace=True) # remove raw features from the dataset
    return df

In [129]:

numeric_features = ['distance']
# numeric_features = ['kills', 'deaths', 'assists', 'denies', 'gold', 'xp', 'health', 'max_health',
#                     'max_mana', 'level', 'towers_killed', 'stuns', 'creeps_stacked', 'camps_stacked',
#                     'lh', 'rune_pickups', 'firstblood_claimed', 'teamfight_participation', 'roshans_killed',
#                     'obs_placed', 'sen_placed']

In [130]:

df_new_features = combine_numeric_features(df_new_features, numeric_features)
test_new_features = combine_numeric_features(test_new_features, numeric_features)

In [70]:

df_new_features.shape, test_new_features.shape

Out[70]:

((79350, 734), (10000, 734))

In [71]:

# import pickle as pkl
# df_new_features.to_pickle('train2.pkl')
# test_new_features.to_pickle('test2.pkl')

In [131]:

df_new_features.head()

Out[131]:

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_kills	r1_deaths	r1_assists	r1_denies	r1_gold	r1_lh	r1_xp	r1_health	r1_max_health	r1_max_mana	r1_level	r1_stuns	r1_creeps_stacked	r1_camps_stacked	r1_rune_pickups	r1_teamfight_participation	r1_obs_placed	r1_ability_level	r1_max_hero_hit	r1_purchase_count	r1_count_ability_use	r1_damage_dealt	r1_damage_received	radiant_tower_kills	dire_tower_kills	diff_tower_kills	r2_hero_id	r2_kills	r2_deaths	r2_assists	r2_denies	r2_gold	r2_lh	r2_xp	r2_health	r2_max_health	r2_max_mana	r2_level	r2_stuns	r2_rune_pickups	r2_firstblood_claimed	r2_teamfight_participation	r2_obs_placed	r2_sen_placed	r2_ability_level	r2_max_hero_hit	r2_purchase_count	r2_count_ability_use	r2_damage_dealt	r2_damage_received	r3_hero_id	r3_kills	r3_deaths	r3_assists	r3_denies	r3_gold	r3_lh	r3_xp	r3_health	r3_max_health	r3_max_mana	r3_level	r3_stuns	r3_creeps_stacked	r3_camps_stacked	r3_rune_pickups	r3_teamfight_participation	r3_obs_placed	r3_sen_placed	r3_ability_level	r3_max_hero_hit	r3_purchase_count	r3_count_ability_use	r3_damage_dealt	r3_damage_received	r4_hero_id	r4_kills	r4_deaths	r4_assists	r4_denies	r4_gold	r4_lh	r4_xp	r4_health	r4_max_health	r4_max_mana	r4_level	r4_rune_pickups	r4_firstblood_claimed	r4_teamfight_participation	r4_obs_placed	r4_sen_placed	r4_ability_level	r4_max_hero_hit	r4_purchase_count	r4_count_ability_use	r4_damage_dealt	r4_damage_received	r5_hero_id	r5_kills	r5_deaths	r5_assists	r5_denies	r5_gold	r5_lh	r5_xp	r5_health	r5_max_health	r5_max_mana	r5_level	r5_stuns	r5_rune_pickups	r5_teamfight_participation	r5_towers_killed	r5_ability_level	r5_max_hero_hit	r5_purchase_count	r5_count_ability_use	r5_damage_dealt	r5_damage_received	d1_hero_id	d1_kills	d1_deaths	d1_assists	d1_denies	d1_gold	d1_lh	d1_xp	d1_health	d1_max_health	d1_max_mana	d1_level	d1_stuns	d1_creeps_stacked	d1_camps_stacked	d1_rune_pickups	d1_teamfight_participation	d1_obs_placed	d1_ability_level	d1_max_hero_hit	d1_purchase_count	d1_count_ability_use	d1_damage_dealt	d1_damage_received	d2_hero_id	d2_kills	d2_deaths	d2_assists	d2_denies	d2_gold	d2_lh	d2_xp	d2_health	d2_max_health	d2_max_mana	d2_level	d2_stuns	d2_rune_pickups	d2_firstblood_claimed	d2_teamfight_participation	d2_obs_placed	d2_sen_placed	d2_ability_level	d2_max_hero_hit	d2_purchase_count	d2_count_ability_use	d2_damage_dealt	d2_damage_received	d3_hero_id	d3_kills	d3_deaths	d3_assists	d3_denies	d3_gold	d3_lh	d3_xp	d3_health	d3_max_health	d3_max_mana	d3_level	d3_stuns	d3_creeps_stacked	d3_camps_stacked	d3_rune_pickups	d3_teamfight_participation	d3_obs_placed	d3_sen_placed	d3_ability_level	d3_max_hero_hit	d3_purchase_count	d3_count_ability_use	d3_damage_dealt	d3_damage_received	d4_hero_id	d4_kills	d4_deaths	d4_assists	d4_denies	d4_gold	d4_lh	d4_xp	d4_health	d4_max_health	d4_max_mana	d4_level	d4_rune_pickups	d4_firstblood_claimed	d4_teamfight_participation	d4_obs_placed	d4_sen_placed	d4_ability_level	d4_max_hero_hit	d4_purchase_count	d4_count_ability_use	d4_damage_dealt	d4_damage_received	d5_hero_id	d5_kills	d5_deaths	d5_assists	d5_denies	d5_gold	d5_lh	d5_xp	d5_health	d5_max_health	d5_max_mana	d5_level	d5_stuns	d5_rune_pickups	d5_teamfight_participation	d5_towers_killed	d5_ability_level	d5_max_hero_hit	d5_purchase_count	d5_count_ability_use	d5_damage_dealt	d5_damage_received	rad_damage_vs_heroes	dire_damage_vs_heroes	bb_restr_Diff	obs_on_map_Diff	rad_damage_vs_tower	dire_damage_vs_tower	damage_vs_tower_Percent	rad_damage_vs_creeps	dire_damage_vs_creeps	damage_vs_creeps_Percent	healing_Diff	healing_Percent	alive_Percent	dead_Percent	actions_Percent	actions_Diff_permin	items_uses_Percent	items_uses_Diff_permin	ability_Percent	ability_Diff_permin	r_item_ancient_janggo	r_item_blade_of_alacrity	r_item_blight_stone	r_item_boots	r_item_branches	r_item_chainmail	r_item_dust	r_item_magic_stick	r_item_magic_wand	r_item_null_talisman	r_item_orb_of_venom	r_item_pers	r_item_phase_boots	r_item_power_treads	r_item_quelling_blade	r_item_recipe_bloodthorn	r_item_recipe_crimson_guard	r_item_recipe_mekansm	r_item_recipe_pipe	r_item_ring_of_aquila	r_item_ring_of_health	r_item_smoke_of_deceit	r_item_soul_ring	r_item_stout_shield	r_item_tango_single	r_item_tranquil_boots	r_item_ward_dispenser	r_item_wind_lace	r_item_wraith_band	d_item_ancient_janggo	d_item_blade_of_alacrity	d_item_blight_stone	d_item_boots	d_item_branches	d_item_chainmail	d_item_circlet	d_item_dust	d_item_magic_stick	d_item_magic_wand	d_item_null_talisman	d_item_orb_of_venom	d_item_pers	d_item_phase_boots	d_item_power_treads	d_item_quelling_blade	d_item_recipe_bloodthorn	d_item_recipe_crimson_guard	d_item_recipe_pipe	d_item_ring_of_aquila	d_item_ring_of_health	d_item_smoke_of_deceit	d_item_soul_ring	d_item_stout_shield	d_item_tango_single	d_item_tranquil_boots	d_item_ward_dispenser	d_item_wind_lace	d_item_wraith_band	r_total_kills	d_total_kills	total_kills_ratio	r_std_kills	d_std_kills	std_kills_ratio	r_total_deaths	d_total_deaths	total_deaths_ratio	r_std_deaths	d_std_deaths	std_deaths_ratio	r_total_assists	d_total_assists	total_assists_ratio	r_std_assists	d_std_assists	std_assists_ratio	r_total_denies	d_total_denies	total_denies_ratio	r_std_denies	d_std_denies	std_denies_ratio	r_total_gold	d_total_gold	total_gold_ratio	r_std_gold	d_std_gold	std_gold_ratio	r_total_lh	d_total_lh	total_lh_ratio	r_std_lh	d_std_lh	std_lh_ratio	r_total_xp	d_total_xp	total_xp_ratio	r_std_xp	d_std_xp	std_xp_ratio	r_total_health	d_total_health	total_health_ratio	r_std_health	d_std_health	std_health_ratio	r_total_max_health	d_total_max_health	total_max_health_ratio	r_std_max_health	d_std_max_health	std_max_health_ratio	r_total_max_mana	d_total_max_mana	total_max_mana_ratio	r_std_max_mana	d_std_max_mana	std_max_mana_ratio	r_total_level	d_total_level	total_level_ratio	r_std_level	d_std_level	std_level_ratio	r_total_x	d_total_x	total_x_ratio	r_std_x	d_std_x	std_x_ratio	r_total_y	d_total_y	total_y_ratio	r_std_y	d_std_y	std_y_ratio	r_total_stuns	d_total_stuns	total_stuns_ratio	r_std_stuns	d_std_stuns	std_stuns_ratio	r_total_creeps_stacked	d_total_creeps_stacked	total_creeps_stacked_ratio	r_std_creeps_stacked	d_std_creeps_stacked	std_creeps_stacked_ratio	r_total_camps_stacked	d_total_camps_stacked	total_camps_stacked_ratio	r_std_camps_stacked	d_std_camps_stacked	std_camps_stacked_ratio	r_total_rune_pickups	d_total_rune_pickups	total_rune_pickups_ratio	r_std_rune_pickups	d_std_rune_pickups	std_rune_pickups_ratio	r_total_firstblood_claimed	d_total_firstblood_claimed	total_firstblood_claimed_ratio	r_std_firstblood_claimed	d_std_firstblood_claimed	std_firstblood_claimed_ratio	r_total_teamfight_participation	d_total_teamfight_participation	total_teamfight_participation_ratio	r_std_teamfight_participation	d_std_teamfight_participation	std_teamfight_participation_ratio	r_total_towers_killed	d_total_towers_killed	total_towers_killed_ratio	r_std_towers_killed	d_std_towers_killed	std_towers_killed_ratio	total_roshans_killed_ratio	std_roshans_killed_ratio	r_total_obs_placed	d_total_obs_placed	total_obs_placed_ratio	r_std_obs_placed	d_std_obs_placed	std_obs_placed_ratio	r_total_sen_placed	d_total_sen_placed	total_sen_placed_ratio	r_std_sen_placed	d_std_sen_placed	std_sen_placed_ratio	r_total_ability_level	d_total_ability_level	total_ability_level_ratio	r_std_ability_level	d_std_ability_level	std_ability_level_ratio	r_total_max_hero_hit	d_total_max_hero_hit	total_max_hero_hit_ratio	r_std_max_hero_hit	d_std_max_hero_hit	std_max_hero_hit_ratio	r_total_purchase_count	d_total_purchase_count	total_purchase_count_ratio	r_std_purchase_count	d_std_purchase_count	std_purchase_count_ratio	r_total_count_ability_use	d_total_count_ability_use	total_count_ability_use_ratio	r_std_count_ability_use	d_std_count_ability_use	std_count_ability_use_ratio	r_total_damage_dealt	d_total_damage_dealt	total_damage_dealt_ratio	r_std_damage_dealt	d_std_damage_dealt	std_damage_dealt_ratio	r_total_damage_received	d_total_damage_received	total_damage_received_ratio	r_std_damage_received	d_std_damage_received	std_damage_received_ratio	r_hero_1	r_hero_11	r_hero_12	r_hero_14	r_hero_15	r_hero_21	r_hero_27	r_hero_34	r_hero_44	r_hero_49	r_hero_51	r_hero_53	r_hero_56	r_hero_58	r_hero_59	r_hero_60	r_hero_63	r_hero_77	r_hero_78	r_hero_84	r_hero_89	r_hero_92	r_hero_96	r_hero_101	d_hero_1	d_hero_11	d_hero_12	d_hero_14	d_hero_15	d_hero_17	d_hero_18	d_hero_21	d_hero_27	d_hero_34	d_hero_40	d_hero_47	d_hero_56	d_hero_58	d_hero_59	d_hero_60	d_hero_63	d_hero_67	d_hero_77	d_hero_78	d_hero_84	d_hero_89	d_hero_92	d_hero_96	r_distance_max	r_distance_min	d_distance_max	d_distance_min
a400b8f29dece5f4d266f49f1ae2e98a	155	22	7	1	11	11	0	0	0	0	543	7	533	358	600	350.938	2	0.000	0	0	1	0.000	0	0	136	6	3	1287	973	0	0	0	78	0	0	0	3	399	4	478	636	720	254.938	2	0.000	0	0	0.000	0	0	0	0	6	2	1327	264	14	0	1	0	0	304	0	130	700	700	242.938	1	0.000	0	0	1	0.000	0	0	0	4	5	1	32	1108	59	0	0	0	1	389	4	506	399	700	326.938	2	0	0	0.000	0	0	0	42	9	0	1000	1192	77	0	0	0	0	402	10	344	422	800	314.938	2	0.000	0	0.000	0	0	0	7	1	5090	705	12	0	0	1	13	982	12	780	650	720	386.938	3	0.000	0	0	1	1.000	0	0	76	7	1	3521	347	21	0	0	0	6	788	9	706	640	640	422.938	3	0.000	2	0	0.000	0	0	0	50	8	0	1977	190	60	0	0	0	1	531	0	307	720	720	242.938	2	0.300	0	0	2	0.000	0	0	0	68	5	3	747	321	84	1	0	0	0	796	0	421	760	760	326.938	2	2	1	1.000	1	0	0	62	9	3	868	16	34	0	0	0	0	851	11	870	593	680	566.938	3	0.000	0	0.000	0	0	164	6	4	2332	681	948.0	2949.0	0.0	-1.0	0.0	0.0	0.500000	2180.0	4090.0	0.347687	-289.0	0.003436	0.498382	1.000000	0.591159	409.666667	0.625000	2.666667	0.388889	-1.333333	0.0	0.0	0.0	0.0	2.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	1.0	NaN	0.0	0.0	NaN	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	2.0	1.0	0.0	0.0	0.0	2.0	3.0	0.0	0.0	0.0	0.0	1.0	1.0	3.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0	1.0	0.0	0.0	0.0	0.0	0	1	0.000000	0.000000	0.447214	0.000000	1	0	inf	0.447214	0.000000	inf	0	1	0.000000	0.000000	0.447214	0.000000	4	20	0.200000	1.303840	5.612486	0.232311	2037	3948	0.515957	85.902852	164.092352	0.523503	25	32	0.781250	3.741657	5.941380	0.629762	1991	3084	0.645590	166.604922	241.358033	0.690281	2515	3363	0.747844	154.029218	66.699325	2.309307	3520	3520	1.000000	71.274119	45.607017	1.562788	1490.69	1946.69	0.765756	46.938257	120.419268	0.389790	9	13	0.692308	0.447214	0.547723	0.816497	600	654	0.917431	35.468296	45.663990	0.776724	590	622	0.948553	26.795522	37.293431	0.718505	0.000	0.300	0.000000	0.000000	0.134164	0.000000	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	2	7	0.285714	0.547723	0.894427	0.612372	0	1	0.0	0.000000	0.447214	0.0	0.000	2.000	0.00	0.000000	0.547723	0.000000	0	0	NaN	0.000000	0.000000	NaN	NaN	NaN	0	1	0.00	0.000000	0.447214	0.000000	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	182	420	0.433333	58.419175	45.716518	1.277857	33	35	0.942857	1.516575	1.581139	0.959166	7	11	0.636364	1.140175	1.643168	0.693889	8736	9445	0.924934	1940.546495	1141.711654	1.699682	4242	1555	2.727974	375.188353	244.929582	1.531821	0	1	0	1	0	0	0	0	0	0	0	0	0	0	1	0	0	1	1	0	0	0	0	0	0	0	1	0	0	0	0	1	0	1	0	0	0	0	0	1	0	0	0	0	1	0	0	0	190.515091	156.204994	198.635344	174.928557
a89e2ea1f94f662d4f5eced92f8b004a	155	22	7	1	11	12	0	0	1	13	982	12	780	650	720	386.938	3	0.000	0	0	1	1.000	0	0	76	7	1	3521	347	0	0	0	21	0	0	0	6	788	9	706	640	640	422.938	3	0.000	2	0	0.000	0	0	0	50	8	0	1977	190	60	0	0	0	1	531	0	307	720	720	242.938	2	0.300	0	0	2	0.000	0	0	0	68	5	3	747	321	84	1	0	0	0	796	0	421	760	760	326.938	2	2	1	1.000	1	0	0	62	9	3	868	16	34	0	0	0	0	851	11	870	593	680	566.938	3	0.000	0	0.000	0	0	164	6	4	2332	681	11	0	0	0	0	543	7	533	358	600	350.938	2	0.000	0	0	1	0.000	0	0	136	6	3	1287	973	78	0	0	0	3	399	4	478	636	720	254.938	2	0.000	0	0	0.000	0	0	0	0	6	2	1327	264	14	0	1	0	0	304	0	130	700	700	242.938	1	0.000	0	0	1	0.000	0	0	0	4	5	1	32	1108	59	0	0	0	1	389	4	506	399	700	326.938	2	0	0	0.000	0	0	0	42	9	0	1000	1192	77	0	0	0	0	402	10	344	422	800	314.938	2	0.000	0	0.000	0	0	0	7	1	5090	705	0.0	0.0	0.0	0.0	0.0	0.0	0.000000	0.0	0.0	0.000000	0.0	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.0	0.0	0.0	2.0	3.0	0.0	0.0	0.0	1.0	1.0	3.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0	0.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	1.0	NaN	0.0	NaN	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	2.0	1.0	1	0	inf	0.447214	0.000000	inf	0	1	0.00	0.000000	0.447214	0.000000	1	0	inf	0.447214	0.000000	inf	20	4	5.000000	5.612486	1.303840	4.304580	3948	2037	1.938144	164.092352	85.902852	1.910208	32	25	1.280000	5.941380	3.741657	1.587901	3084	1991	1.548970	241.358033	166.604922	1.448685	3363	2515	1.337177	66.699325	154.029218	0.433030	3520	3520	1.000000	45.607017	71.274119	0.639882	1946.69	1490.69	1.305899	120.419268	46.938257	2.565482	13	9	1.444444	0.547723	0.447214	1.224745	654	600	1.090000	45.663990	35.468296	1.287459	622	590	1.054237	37.293431	26.795522	1.391778	0.300	0.000	inf	0.134164	0.000000	inf	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	7	2	3.500000	0.894427	0.547723	1.632993	1	0	inf	0.447214	0.000000	inf	2.000	0.000	inf	0.547723	0.000000	inf	0	0	NaN	0.000000	0.000000	NaN	NaN	NaN	1	0	inf	0.447214	0.000000	inf	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	420	182	2.307692	45.716518	58.419175	0.782560	35	33	1.060606	1.581139	1.516575	1.042572	11	7	1.571429	1.643168	1.140175	1.441153	9445	8736	1.081158	1141.711654	1940.546495	0.588345	1555	4242	0.366572	244.929582	375.188353	0.652818	0	0	1	0	0	1	0	1	0	0	0	0	0	0	0	1	0	0	0	1	0	0	0	0	0	1	0	1	0	0	0	0	0	0	0	0	0	0	1	0	0	0	1	1	0	0	0	0	198.635344	174.928557	190.515091	156.204994
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	15	7	2	0	7	5257	52	3937	1160	1160	566.938	8	0.000	0	0	0	0.438	0	6	216	28	9	20882	4857	2	0	2	96	3	1	2	3	3394	19	3897	1352	1380	386.938	8	8.398	4	0	0.312	0	0	6	247	19	21	14983	7114	27	1	1	4	2	2212	4	2561	710	860	530.938	6	11.965	2	1	4	0.312	3	1	4	153	27	11	5941	1758	63	4	0	3	12	4206	38	4459	420	880	482.938	9	3	0	0.438	1	2	7	201	21	35	21394	5410	89	1	0	5	4	3103	14	2712	856	900	446.938	6	21.697	2	0.375	1	5	152	9	18	6565	2064	58	1	2	0	4	2823	24	3281	700	700	686.938	7	3.166	1	1	3	0.250	1	4	114	15	11	13575	3096	14	1	6	0	1	2466	17	2360	758	1040	326.938	6	0.067	1	1	0.250	4	2	4	182	19	45	9643	7316	1	1	3	1	7	3624	29	3418	485	800	350.938	7	0.300	2	1	4	0.500	0	0	5	226	20	22	13855	6000	56	0	3	2	3	2808	18	2730	567	1160	410.938	6	6	0	0.500	0	0	5	78	23	17	7437	5893	92	0	2	0	1	1423	8	1136	800	800	446.938	4	0.000	0	0.000	0	4	164	11	7	2308	2154	17463.0	13063.0	-1.0	-2.0	1970.0	12.0	0.993946	41940.0	12526.0	0.770022	-99.0	0.486901	0.509275	0.290393	0.724752	1016.818182	0.605882	3.272727	0.479592	-0.727273	1.0	1.0	1.0	2.0	0.0	1.0	2.0	0.0	3.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	1.0	2.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	1.0	2.0	0.0	0.0	0.0	0.0	2.0	1.0	0.0	NaN	0.0	0.0	1.0	1.0	0.0	1.0	0.0	1.0	0.0	0.0	1.0	16	3	5.333333	2.489980	0.547723	4.546061	4	16	0.25	0.836660	1.643168	0.509175	14	3	4.666667	1.923538	0.894427	2.150581	28	16	1.750000	4.037326	2.489980	1.621429	18172	13144	1.382532	1153.576742	796.979736	1.447435	127	96	1.322917	19.333908	7.918333	2.441664	17566	12925	1.359072	832.173780	915.089613	0.909390	4498	3310	1.358912	367.538841	132.342359	2.777182	5180	4500	1.151111	227.771816	191.833261	1.187343	2414.69	2222.69	1.086382	70.484041	143.599443	0.490838	37	30	1.233333	1.341641	1.224745	1.095445	614	672	0.913690	41.871231	35.054244	1.194470	686	730	0.939726	34.076385	30.822070	1.105584	42.060	3.533	11.904897	9.091737	1.380365	6.586474	2	3	0.666667	0.894427	0.894427	1.0	1	2	0.5	0.447214	0.547723	0.816497	13	14	0.928571	1.673320	2.387467	0.700877	0	1	0.0	0.000000	0.447214	0.0	1.875	1.500	1.25	0.063000	0.209165	0.301198	1	0	inf	0.447214	0.000000	inf	NaN	NaN	4	5	0.80	1.303840	1.732051	0.752773	3	2	1.500000	0.894427	0.894427	1.0	28	22	1.272727	1.140175	0.547723	2.081666	969	764	1.268325	41.191018	57.958606	0.710697	104	88	1.181818	7.628892	4.669047	1.633929	94	102	0.921569	10.305338	14.892951	0.691961	69765	46818	1.490132	7470.192601	4782.256455	1.562064	21203	24459	0.866879	2286.086350	2169.544929	1.053717	0	0	0	0	1	0	1	0	0	0	0	0	0	0	0	0	1	0	0	0	1	0	1	0	1	0	0	1	0	0	0	0	0	0	0	0	1	1	0	0	0	0	0	0	0	0	1	0	213.663287	108.903627	251.745904	184.228119
d441caf69ec9c97fa2a47ec054c75c9b	658	4	0	3	10	58	1	2	0	4	2823	24	3281	700	700	686.938	7	3.166	1	1	3	0.250	1	4	114	15	11	13575	3096	0	2	-2	14	1	6	0	1	2466	17	2360	758	1040	326.938	6	0.067	1	1	0.250	4	2	4	182	19	45	9643	7316	1	1	3	1	7	3624	29	3418	485	800	350.938	7	0.300	2	1	4	0.500	0	0	5	226	20	22	13855	6000	56	0	3	2	3	2808	18	2730	567	1160	410.938	6	6	0	0.500	0	0	5	78	23	17	7437	5893	92	0	2	0	1	1423	8	1136	800	800	446.938	4	0.000	0	0.000	0	4	164	11	7	2308	2154	15	7	2	0	7	5257	52	3937	1160	1160	566.938	8	0.000	0	0	0	0.438	0	6	216	28	9	20882	4857	96	3	1	2	3	3394	19	3897	1352	1380	386.938	8	8.398	4	0	0.312	0	0	6	247	19	21	14983	7114	27	1	1	4	2	2212	4	2561	710	860	530.938	6	11.965	2	1	4	0.312	3	1	4	153	27	11	5941	1758	63	4	0	3	12	4206	38	4459	420	880	482.938	9	3	0	0.438	1	2	7	201	21	35	21394	5410	89	1	0	5	4	3103	14	2712	856	900	446.938	6	21.697	2	0.375	1	5	152	9	18	6565	2064	0.0	0.0	0.0	0.0	0.0	0.0	0.000000	0.0	0.0	0.000000	0.0	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.0	0.0	0.0	1.0	1.0	0.0	0.0	1.0	2.0	0.0	0.0	0.0	0.0	2.0	1.0	0.0	NaN	NaN	0.0	0.0	1.0	1.0	0.0	1.0	0.0	1.0	0.0	0.0	1.0	1.0	1.0	1.0	2.0	0.0	1.0	0.0	2.0	0.0	3.0	0.0	0.0	1.0	1.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	1.0	2.0	0.0	1.0	1.0	0.0	0.0	3	16	0.187500	0.547723	2.489980	0.219971	16	4	4.00	1.643168	0.836660	1.963961	3	14	0.214286	0.894427	1.923538	0.464991	16	28	0.571429	2.489980	4.037326	0.616740	13144	18172	0.723311	796.979736	1153.576742	0.690877	96	127	0.755906	7.918333	19.333908	0.409557	12925	17566	0.735796	915.089613	832.173780	1.099638	3310	4498	0.735883	132.342359	367.538841	0.360077	4500	5180	0.868726	191.833261	227.771816	0.842217	2222.69	2414.69	0.920487	143.599443	70.484041	2.037333	30	37	0.810811	1.224745	1.341641	0.912871	672	614	1.094463	35.054244	41.871231	0.837192	730	686	1.064140	30.822070	34.076385	0.904499	3.533	42.060	0.083999	1.380365	9.091737	0.151826	3	2	1.500000	0.894427	0.894427	1.0	2	1	2.0	0.547723	0.447214	1.224745	14	13	1.076923	2.387467	1.673320	1.426785	1	0	inf	0.447214	0.000000	inf	1.500	1.875	0.80	0.209165	0.063000	3.320079	0	1	0.0	0.000000	0.447214	0.0	NaN	NaN	5	4	1.25	1.732051	1.303840	1.328422	2	3	0.666667	0.894427	0.894427	1.0	22	28	0.785714	0.547723	1.140175	0.480384	764	969	0.788442	57.958606	41.191018	1.407069	88	104	0.846154	4.669047	7.628892	0.612022	102	94	1.085106	14.892951	10.305338	1.445169	46818	69765	0.671081	4782.256455	7470.192601	0.640178	24459	21203	1.153563	2169.544929	2286.086350	0.949021	1	0	0	1	0	0	0	0	0	0	0	0	1	1	0	0	0	0	0	0	0	1	0	0	0	0	0	0	1	0	0	0	1	0	0	0	0	0	0	0	1	0	0	0	0	1	0	1	251.745904	184.228119	213.663287	108.903627
6db558535151ea18ca70a6892197db41	21	23	0	0	0	101	0	0	0	0	176	0	0	680	680	506.938	1	0.000	0	0	0	0.000	0	0	0	4	0	0	0	0	0	0	51	0	0	0	0	176	0	0	720	720	278.938	1	0.000	0	0	0.000	0	0	0	0	7	0	0	0	44	0	0	0	0	176	0	0	568	600	254.938	1	0.000	0	0	1	0.000	0	0	0	70	3	1	70	68	49	0	0	0	0	176	0	0	580	580	254.938	1	1	0	0.000	0	0	0	0	4	0	0	0	53	0	0	0	0	176	0	0	580	580	374.938	1	0.000	1	0.000	0	0	0	3	0	0	0	18	0	0	0	0	96	0	0	660	660	266.938	1	0.000	0	0	0	0.000	0	0	0	1	0	0	0	67	0	0	0	0	96	0	0	586	620	278.938	1	0.000	0	0	0.000	0	0	0	68	5	1	68	70	47	0	0	0	0	96	0	0	660	660	290.938	1	0.000	0	0	1	0.000	0	0	0	0	3	0	0	0	40	0	0	0	0	96	0	0	600	600	302.938	1	0	0	0.000	0	0	0	0	5	0	0	0	17	0	0	0	0	96	0	0	640	640	446.938	1	0.000	0	0.000	0	0	0	4	0	0	0	70.0	68.0	0.0	0.0	0.0	0.0	0.500000	0.0	0.0	0.500000	0.0	0.500000	0.500000	0.500000	0.500000	0.000000	0.000000	-1.000000	0.500000	0.000000	0.0	0.0	1.0	0.0	2.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	3.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	1.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	2.0	0.0	0.0	0.0	0.0	1.0	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	880	480	1.833333	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	3128	3146	0.994278	69.547106	34.397674	2.021855	3160	3180	0.993711	64.187226	26.076810	2.461468	1670.69	1586.69	1.052940	108.532023	73.680391	1.473011	5	5	1.000000	0.000000	0.000000	NaN	580	796	0.728643	37.576588	33.840804	1.110393	586	736	0.796196	27.589853	33.573799	0.821767	0.000	0.000	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	3	1	3.000000	0.547723	0.447214	1.224745	0	0	NaN	0.000000	0.000000	NaN	0.000	0.000	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	NaN	NaN	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	0	0	NaN	0.000000	0.000000	NaN	70	68	1.029412	31.304952	30.410524	1.029412	21	18	1.166667	1.643168	1.673320	0.981981	1	1	1.000000	0.447214	0.447214	1.000000	70	68	1.029412	31.304952	30.410524	1.029412	68	70	0.971429	30.410524	31.304952	0.971429	0	0	0	0	0	0	0	0	1	1	1	1	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	1	1	0	0	0	1	1	0	0	0	0	0	1	0	0	0	0	0	0	187.488666	162.012345	253.148178	200.688814

In [73]:

# X = df_new_features.reset_index(drop=True)
# X_test = test_new_features.copy().reset_index(drop=True)

In [132]:

X.shape, X_test.shape

Out[132]:

((79350, 734), (10000, 734))

In [133]:

%%time
X = df_new_features.reset_index(drop=True)
X_test = test_new_features.copy().reset_index(drop=True)

oof_lgb, prediction_lgb, scores = train_model(X, X_test, y_train_inverse, params=params2, folds=folds, model_type='lgb', plot_feature_importance=True)

Fold 0 started at Tue Nov 19 11:13:55 2019
Training until validation scores don't improve for 200 rounds.
[1000]	training's auc: 0.861552	valid_1's auc: 0.846879
[2000]	training's auc: 0.882164	valid_1's auc: 0.850581
[3000]	training's auc: 0.898479	valid_1's auc: 0.851424
Early stopping, best iteration is:
[3078]	training's auc: 0.899768	valid_1's auc: 0.85158
Fold 1 started at Tue Nov 19 11:16:45 2019
Training until validation scores don't improve for 200 rounds.
[1000]	training's auc: 0.861904	valid_1's auc: 0.844829
[2000]	training's auc: 0.882257	valid_1's auc: 0.848995
[3000]	training's auc: 0.898637	valid_1's auc: 0.850187
[4000]	training's auc: 0.912741	valid_1's auc: 0.850578
Early stopping, best iteration is:
[4063]	training's auc: 0.913571	valid_1's auc: 0.850588
Fold 2 started at Tue Nov 19 11:20:11 2019
Training until validation scores don't improve for 200 rounds.
[1000]	training's auc: 0.861746	valid_1's auc: 0.845504
[2000]	training's auc: 0.882032	valid_1's auc: 0.849496
[3000]	training's auc: 0.898449	valid_1's auc: 0.850742
[4000]	training's auc: 0.912855	valid_1's auc: 0.85123
Early stopping, best iteration is:
[3843]	training's auc: 0.910684	valid_1's auc: 0.851285
Fold 3 started at Tue Nov 19 11:23:26 2019
Training until validation scores don't improve for 200 rounds.
[1000]	training's auc: 0.861664	valid_1's auc: 0.846237
[2000]	training's auc: 0.88209	valid_1's auc: 0.850764
[3000]	training's auc: 0.898453	valid_1's auc: 0.851932
Early stopping, best iteration is:
[3717]	training's auc: 0.908748	valid_1's auc: 0.852411
Fold 4 started at Tue Nov 19 11:26:36 2019
Training until validation scores don't improve for 200 rounds.
[1000]	training's auc: 0.861871	valid_1's auc: 0.845653
[2000]	training's auc: 0.882321	valid_1's auc: 0.849698
[3000]	training's auc: 0.898629	valid_1's auc: 0.85117
[4000]	training's auc: 0.912781	valid_1's auc: 0.851666
Early stopping, best iteration is:
[3952]	training's auc: 0.912177	valid_1's auc: 0.851691
CV mean score: 0.8515, std: 0.0006.
CPU times: user 58min 55s, sys: 1min 26s, total: 1h 22s
Wall time: 16min

In [132]:

sub = pd.read_csv('/Users/mant1/input/dota_csv/sample_submission.csv')
sub['radiant_win_prob'] = prediction_lgb
sub.to_csv('submission_v15.csv', index=False)
sub.head()

Out[132]:

	match_id_hash	radiant_win_prob
0	30cc2d778dca82f2edb568ce9b585caa	0.284332
1	70e5ba30f367cea48793b9003fab9d38	0.992162
2	4d9ef74d3a2025d79e9423105fd73d41	0.975972
3	2bb79e0c1eaac1608e5a09c8e0c6a555	0.827566
4	bec17f099b01d67edc82dfb5ce735a43	0.274927

In [141]:

plt.hist(prediction_lgb, bins=40);
plt.title('Distribution of predictions');