This notebook explains how to generate feature importance plots from scikit-learn
using tree-based feature importance, permutation importance and shap
.
This notebook will build and evaluate a model to predict arrival delay for flights in and out of NYC in 2013.
import statsmodels.api as sm
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import train_test_split
from sklearn.inspection import permutation_importance
import shap
import category_encoders as ce
from sklearn.ensemble import GradientBoostingRegressor
The data is from rdatasets
imported using the Python package statsmodels
.
df = sm.datasets.get_rdataset('flights', 'nycflights13').data
df.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 336776 entries, 0 to 336775 Data columns (total 19 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 year 336776 non-null int64 1 month 336776 non-null int64 2 day 336776 non-null int64 3 dep_time 328521 non-null float64 4 sched_dep_time 336776 non-null int64 5 dep_delay 328521 non-null float64 6 arr_time 328063 non-null float64 7 sched_arr_time 336776 non-null int64 8 arr_delay 327346 non-null float64 9 carrier 336776 non-null object 10 flight 336776 non-null int64 11 tailnum 334264 non-null object 12 origin 336776 non-null object 13 dest 336776 non-null object 14 air_time 327346 non-null float64 15 distance 336776 non-null int64 16 hour 336776 non-null int64 17 minute 336776 non-null int64 18 time_hour 336776 non-null object dtypes: float64(5), int64(9), object(5) memory usage: 48.8+ MB
df.isnull().sum()
year 0 month 0 day 0 dep_time 8255 sched_dep_time 0 dep_delay 8255 arr_time 8713 sched_arr_time 0 arr_delay 9430 carrier 0 flight 0 tailnum 2512 origin 0 dest 0 air_time 9430 distance 0 hour 0 minute 0 time_hour 0 dtype: int64
As this model will predict arrival delay, the Null
values are caused by flights did were cancelled or diverted. These can be excluded from this analysis.
df.dropna(inplace=True)
df['arr_hour'] = df.arr_time.apply(lambda x: int(np.floor(x/100)))
df['arr_minute'] = df.arr_time.apply(lambda x: int(x - np.floor(x/100)*100))
df['sched_arr_hour'] = df.sched_arr_time.apply(lambda x: int(np.floor(x/100)))
df['sched_arr_minute'] = df.sched_arr_time.apply(lambda x: int(x - np.floor(x/100)*100))
df['sched_dep_hour'] = df.sched_dep_time.apply(lambda x: int(np.floor(x/100)))
df['sched_dep_minute'] = df.sched_dep_time.apply(lambda x: int(x - np.floor(x/100)*100))
df.rename(columns={'hour': 'dep_hour',
'minute': 'dep_minute'}, inplace=True)
target = 'arr_delay'
y = df[target]
X = df.drop(columns=[target, 'flight', 'tailnum', 'time_hour', 'year', 'dep_time', 'sched_dep_time', 'arr_time', 'sched_arr_time', 'dep_delay'])
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2, random_state=1066)
We use a leave-one-out encoder as it creates a single column for each categorical variable instead of creating a column for each level of the categorical variable like one-hot-encoding. This makes interpreting the impact of categorical variables with feature impact easier.
encoder = ce.LeaveOneOutEncoder(return_df=True)
X_train_loo = encoder.fit_transform(X_train, y_train)
X_test_loo = encoder.transform(X_test)
model = GradientBoostingRegressor(learning_rate=0.05, max_depth=5, n_estimators=500, min_samples_split=5, n_iter_no_change=10)
model.fit(X_train_loo, y_train)
rmse = np.sqrt(mean_squared_error(y_test, model.predict(X_test_loo)))
rmse
43.000748477245494
feature_importance = model.feature_importances_
sorted_idx = np.argsort(feature_importance)
fig = plt.figure(figsize=(12, 6))
plt.barh(range(len(sorted_idx)), feature_importance[sorted_idx], align='center')
plt.yticks(range(len(sorted_idx)), np.array(X_test.columns)[sorted_idx])
plt.title('Feature Importance')
Text(0.5, 1.0, 'Feature Importance')
perm_importance = permutation_importance(model, X_test_loo, y_test, n_repeats=10, random_state=1066)
sorted_idx = perm_importance.importances_mean.argsort()
fig = plt.figure(figsize=(12, 6))
plt.barh(range(len(sorted_idx)), perm_importance.importances_mean[sorted_idx], align='center')
plt.yticks(range(len(sorted_idx)), np.array(X_test.columns)[sorted_idx])
plt.title('Permutation Importance')
Text(0.5, 1.0, 'Permutation Importance')
explainer = shap.Explainer(model)
shap_values = explainer(X_test_loo)
shap_importance = shap_values.abs.mean(0).values
sorted_idx = shap_importance.argsort()
fig = plt.figure(figsize=(12, 6))
plt.barh(range(len(sorted_idx)), shap_importance[sorted_idx], align='center')
plt.yticks(range(len(sorted_idx)), np.array(X_test.columns)[sorted_idx])
plt.title('SHAP Importance')
Text(0.5, 1.0, 'SHAP Importance')
SHAP
contains a function to plot this directly.
shap.plots.bar(shap_values, max_display=X_test_loo.shape[0])