# Machine Learning Using Python (MEAFA Workshop)

## Lesson 10: Neural Networks (Regression)

This notebook relies on the following imports and setting. We will load new functions and libraries in context to make clear what we are using them for.

In [1]:
# Packages
import numpy as np
from scipy import stats
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import warnings
warnings.filterwarnings('ignore') # this is to clear the warnings from this page, usually we should leave them on

In [2]:
# Plot settings
sns.set_context('notebook') # optimise figures for notebook display
sns.set_style('ticks') # set default plot style
colours = ['#1F77B4', '#FF7F0E', '#2CA02C', '#DB2728', '#9467BD', '#8C564B', '#E377C2','#7F7F7F', '#BCBD22', '#17BECF']
crayon = ['#4E79A7','#F28E2C','#E15759','#76B7B2','#59A14F', '#EDC949','#AF7AA1','#FF9DA7','#9C755F','#BAB0AB']
sns.set_palette(colours) # set custom color scheme
%matplotlib inline
plt.rcParams['figure.figsize'] = (9, 6)

In [3]:
# Methods
from sklearn.linear_model import LinearRegression

# Model selection and evaluation tools
from sklearn.model_selection import train_test_split
from sklearn.model_selection import GridSearchCV, RandomizedSearchCV
from sklearn.model_selection import cross_val_score
from sklearn.metrics import mean_squared_error, r2_score,  mean_absolute_error

# Data processing
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler


## Credit Card Data¶

In [4]:
data=pd.read_csv('Datasets/Credit.csv', index_col='Obs')

Out[4]:
Income Limit Rating Cards Age Education Gender Student Married Ethnicity Balance
Obs
1 14.891 3606 283 2 34 11 Male No Yes Caucasian 333
2 106.025 6645 483 3 82 15 Female Yes Yes Asian 903
3 104.593 7075 514 4 71 11 Male No No Asian 580
4 148.924 9504 681 3 36 11 Female No No Asian 964
5 55.882 4897 357 2 68 16 Male No Yes Caucasian 331
6 80.180 8047 569 4 77 10 Male No No Caucasian 1151
7 20.996 3388 259 2 37 12 Female No No African American 203
8 71.408 7114 512 2 87 9 Male No No Asian 872
9 15.125 3300 266 5 66 13 Female No No Caucasian 279
10 71.061 6819 491 3 41 19 Female Yes Yes African American 1350
In [5]:
from sklearn.model_selection import train_test_split

# Randomly split indexes
index_train, index_test  = train_test_split(np.array(data.index), train_size=0.7, random_state=10)

# Write training and test sets
train = data.loc[index_train,:].copy()
test =  data.loc[index_test,:].copy()

In [6]:
def prepare_data(df):
df['Male']=(df['Gender'] ==' Male').astype(int) # create dummy variable for gender
df['Student']=(df['Student'] =='Yes').astype(int)
df['Married']=(df['Married'] =='Yes').astype(int)
df['Caucasian']=(df['Ethnicity'] =='Caucasian').astype(int)
df['Asian']=(df['Ethnicity'] =='Asian').astype(int)
df=df.loc[:, df.dtypes!='object'] # discards the columns that are not numerical
df=df.drop('Rating', axis=1) # collinear with limit
return df

train = prepare_data(train)
test = prepare_data(test)


Out[6]:
Income Limit Cards Age Education Student Married Balance Male Caucasian Asian
Obs
400 18.701 5524 5 64 7 0 0 966 0 0 1
26 14.090 4323 5 25 16 0 1 671 0 0 0
280 54.319 3063 3 59 8 1 0 269 0 1 0
261 67.937 5184 4 63 12 0 1 345 1 0 1
131 23.793 3821 4 56 12 1 1 868 0 0 0
In [7]:
# Construting response vector and design matrix (matrix of predictor values)
response = 'Balance'
predictors = list(train.columns.values)
predictors.remove(response)

y_train = train[response].copy()
y_test = test[response].copy()

X_train=train[predictors].copy() # selects the variables in the predictor list

scaler = StandardScaler()
scaler.fit(X_train)
X_train = scaler.transform(X_train)
X_test = scaler.transform(test[predictors].values)

X_test.shape

Out[7]:
(120, 10)

## Neural Networks¶

In [8]:
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.layers import Activation
from keras.wrappers.scikit_learn import KerasRegressor

Using TensorFlow backend.


### Single Layer Perceptron¶

In [9]:
slp = Sequential()
slp.fit(X_train, y_train, epochs=15000, verbose=0)

Out[9]:
<keras.callbacks.History at 0x249c412a780>

### Using the Scikit-Learn Wrapper¶

In [10]:
def build_model():
model = Sequential()
return model

estimator = KerasRegressor(build_fn=build_model, epochs=1000, verbose=0)
cross_val_score(estimator, X_train, y_train, cv=5, scoring = 'neg_mean_squared_error')

Out[10]:
array([-16313.7029001 , -15914.1558259 , -14988.69404694, -29202.73997519,
-16337.20212685])

### Dropout¶

In [11]:
nn = Sequential()
nn.fit(X_train, y_train, epochs=100, verbose=0) # few epochs just for illustration

Out[11]:
<keras.callbacks.History at 0x249c6af3f98>

## Model evaluation¶

In [12]:
# Benchmark
ols = LinearRegression()
ols.fit(X_train, y_train)

# Initialise table
columns=['RMSE', 'R-Squared', 'MAE']
rows=['Linear Regression', 'Neural ']
results =pd.DataFrame(0.0, columns=columns, index=rows)

# List algorithms
methods = [ols, slp]

# Computer test predictions and metrics
for i, method in enumerate(methods):

y_pred = method.predict(X_test)
results.iloc[i, 0] = np.sqrt(mean_squared_error(y_test, y_pred))
results.iloc[i, 1] = r2_score(y_test, y_pred)
results.iloc[i, 2] = mean_absolute_error(y_test, y_pred)

results.round(2)

Out[12]:
RMSE R-Squared MAE
Linear Regression 97.19 0.96 80.03
Neural 11.50 1.00 8.45