STEP 1: Import ALL the things!¶
In [1]:
%tensorflow_version 2.x
from __future__ import absolute_import, division, print_function, unicode_literals
import tensorflow as tf
from tensorflow import keras
import os
import tempfile
import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
import sklearn
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
mpl.rcParams['figure.figsize'] = (12, 10)
colors = plt.rcParams['axes.prop_cycle'].by_key()['color']
(A) With Fraud Data¶
In [2]:
file = tf.keras.utils
raw_df = pd.read_csv('https://storage.googleapis.com/download.tensorflow.org/data/creditcard.csv')
raw_df.head()
Out[2]:
(B) With Loan Data¶
In [3]:
from google.colab import drive
drive.mount('/content/drive')
In [0]:
import os
os.chdir("drive/My Drive/data")
In [0]:
raw_df_B = pd.read_csv('home-credit-default-risk/application_train.csv')
STEP 2: Examine Class Imbalance¶
In [8]:
### A
neg, pos = np.bincount(raw_df['Class'])
total = neg + pos
print('Examples:\n Total: {}\n Positive: {} ({:.2f}% of total)\n'.format(
total, pos, 100 * pos / total))
In [11]:
### B
repaid, defaulted = np.bincount(raw_df_B['TARGET'])
total = repaid + defaulted
print('Examples:\n Total: {}\n Defaulted: {} ({:.2f}% of total)\n'.format(
total, defaulted, 100 * defaulted / total))
STEP 3: Clean, split and normalize the data¶
In [0]:
### A
# You don't want the `Time` column.
# The `Amount` column covers a huge range. Convert to log-space.
cleaned_df = raw_df.copy()
cleaned_df.pop('Time')
eps=0.001 # 0 => 0.1¢
cleaned_df['Log Ammount'] = np.log(cleaned_df.pop('Amount')+eps)
In [0]:
### B
# Temp "cleaning"
cleaned_df_B = raw_df_B.copy()
cols = ['TARGET','CNT_FAM_MEMBERS',
'CNT_CHILDREN',
'AMT_REQ_CREDIT_BUREAU_YEAR',
'OWN_CAR_AGE',
'DAYS_REGISTRATION',
'DAYS_ID_PUBLISH',
'DAYS_LAST_PHONE_CHANGE',
'DAYS_BIRTH','AMT_INCOME_TOTAL']
cleaned_df_B = pd.DataFrame(cleaned_df_B, columns=cols)
In [51]:
print(cleaned_df.shape)
print(cleaned_df_B.shape)
In [0]:
### A
def get_split_data(df, target_label):
# Use a utility from sklearn to split and shuffle our dataset.
train_df, test_df = train_test_split(df, test_size=0.2)
train_df, val_df = train_test_split(train_df, test_size=0.2)
# Form np arrays of labels and features.
train_labels = np.array(train_df.pop(target_label))
bool_train_labels = train_labels != 0
val_labels = np.array(val_df.pop(target_label))
test_labels = np.array(test_df.pop(target_label))
train_features = np.array(train_df)
val_features = np.array(val_df)
test_features = np.array(test_df)
return train_df, train_labels, bool_train_labels, val_labels, test_labels, train_features, val_features, test_features
In [0]:
train_df, train_labels, bool_train_labels, val_labels, test_labels, train_features, val_features, test_features = get_split_data(cleaned_df, 'Class')
In [0]:
train_df_B, train_labels_B, bool_train_labels_B, val_labels_B, test_labels_B, train_features_B, val_features_B, test_features_B = get_split_data(cleaned_df_B, 'TARGET')
In [0]:
def scale_and_plot(train_df, train_labels, val_labels, test_labels, train_features, val_features, test_features, bool_train_labels, col1, col2):
scaler = StandardScaler()
train_features = scaler.fit_transform(train_features)
val_features = scaler.transform(val_features)
test_features = scaler.transform(test_features)
train_features = np.clip(train_features, -5, 5)
val_features = np.clip(val_features, -5, 5)
test_features = np.clip(test_features, -5, 5)
print('Training labels shape:', train_labels.shape)
print('Validation labels shape:', val_labels.shape)
print('Test labels shape:', test_labels.shape)
print('Training features shape:', train_features.shape)
print('Validation features shape:', val_features.shape)
print('Test features shape:', test_features.shape)
pos_df = pd.DataFrame(train_features[ bool_train_labels], columns = train_df.columns)
neg_df = pd.DataFrame(train_features[~bool_train_labels], columns = train_df.columns)
sns.jointplot(pos_df[col1], pos_df[col2],
kind='hex', xlim = (-5,5), ylim = (-5,5))
plt.suptitle("Positive distribution")
sns.jointplot(neg_df[col1], neg_df[col2],
kind='hex', xlim = (-5,5), ylim = (-5,5))
_ = plt.suptitle("Negative distribution")
In [56]:
scale_and_plot(train_df, train_labels, val_labels, test_labels, train_features, val_features, test_features, bool_train_labels, 'V5', 'V6')
In [57]:
scale_and_plot(train_df_B, train_labels_B, val_labels_B, test_labels_B, train_features_B, val_features_B, test_features_B, bool_train_labels_B, 'AMT_INCOME_TOTAL','DAYS_BIRTH')
In [0]: