Churn dataset Image generation code

# -*- coding: utf-8 -*-

"""

Created on Wed Jan 6 23:09:40 2021

@author: Rajee

"""

# -*- coding: utf-8 -*-

"""

Created on Sat Nov 28 23:43:13 2020

@author: Rajeev

"""

import os

os.environ["CUDA_VISIBLE_DEVICES"] = "0" #force cuda

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

import pandas as pd

from tensorflow.keras.utils import to_categorical

import numpy as np

from PIL import Image #image generation, saving

import time

#import churn data

train_csv = pd.read_csv("customer_churn_train.csv", header="infer")

train_x= train_csv.iloc[:,1:20]

train_y= train_csv.loc[:,"Churn?"]

# compute the number of labels

num_labels = 2 # could be inferred from train_y, bad practice alert!

#Change Churn to Numeric binary lables and change other boolean types too

train_y.replace(to_replace="True.",value="1",inplace=True)

train_y.replace(to_replace="False.",value="0",inplace=True)

train_y = to_categorical(train_y) #took me time to figure out why my code was vitriolic. This is necesssary!

train_x.replace(to_replace="yes",value="1",inplace=True)

train_x.replace(to_replace="no",value="0",inplace=True)

#drop phone no column. Add-on exercise, impute missing values or derive features from these columns

train_x.drop(columns='Phone',inplace=True)

train_x.drop(columns="Int'l Plan",inplace=True)

train_x.drop(columns="VMail Plan",inplace=True)

(rows,input_size) = train_x.shape #input layer shape is no of columns

train_x=((train_x-train_x.std())/train_x.mean())

train_x=train_x.abs()

train_x= train_x*256

train_x = train_x.astype(int)

train_x=train_x.clip(lower=0,upper=255)

#image array creation params

nrows =train_x.shape[0] # no of images to generate

w=train_x.shape[1] #no of features is width

h=train_x.shape[1]# lets keep img height same as width

pix=(h,w,3)

myImageArray = np.zeros(pix,dtype=np.uint8)

for i in range(nrows): # each iteration represents a new image

  for j in range(h):

    for k in range(w):

      myImageArray[j,k]=[train_x.iloc[i,k],train_x.iloc[i,k],train_x.iloc[i,k]]

  myI = Image.fromarray(myImageArray,"RGB") #indent this and next row after first For loop

  fp = "telecom-churn-row-" + str(i) + "."+ ".png" #dynamic filename for saving images per row

  myI.save(fp)