digit_recognizer.py

# -*- coding: utf-8 -*-
"""Digit Recognizer.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1OxPs_9gIqc57Cgj7aG0OWgQCX-KPxU_w
"""

from google.colab import drive
drive.mount('/content/drive/')

!ls "/content/drive/My Drive/Colab Notebooks"

!unzip -q "/content/drive/My Drive/digit-recognizer.zip"

import numpy as np
import pandas as pd
import keras
from keras.models import Model
from keras.layers import *
from keras import optimizers
import tensorflow as tf

df_train = pd.read_csv('/content/drive/My Drive/train.csv')
df_test = pd.read_csv('/content/drive/My Drive/test.csv')

print(df_train.shape)
#42000 rows and 785 columns
print(df_test.shape)
#28000 rows in test

df_train.head()
#784 features and 1 label

#iloc The iloc indexer for Pandas Dataframe is used for integer-location based indexing / selection by position
#iloc basically used for dataframe slicing
df_features = df_train.iloc[:, 1:785]
print(df_features.shape)

#extracting al lables tha contains our actual data
df_label = df_train.iloc[:, 0]
print(df_label.shape)

X_test = df_test.iloc[:, 0:784]

print(X_test.shape)

#splitting data into train and test
from sklearn.model_selection import train_test_split
X_train,X_val,y_train,y_val=train_test_split(df_features,df_label,
                                          test_size=0.2,
                                          random_state=1212)

X_train = X_train.as_matrix().reshape(33600, 784) #(33600, 784)
X_val = X_val.as_matrix().reshape(8400, 784) #(8400, 784)

X_test = X_test.as_matrix().reshape(28000, 784)

X_train=tf.keras.utils.normalize(X_train,axis=1)
X_test=tf.keras.utils.normalize(X_test,axis=1)
print((min(X_train[1]), max(X_train[1])))
#normalizing the values same as above code

#convert labels to one hot encoders
num_digits = 10
#Converts a class vector (integers) to binary class matrix.
y_train = keras.utils.to_categorical(y_train, num_digits)
y_val = keras.utils.to_categorical(y_val, num_digits)

print(y_train[2])

#Model fitting 
#building a neural network of 5 layers
model=tf.keras.models.Sequential()
#creating input layer
#model.add(tf.keras.layers.Flatten())

#hidden layers creation
model.add(tf.keras.layers.Dense(300,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(200,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(200,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(200,activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(100,activation=tf.nn.relu))

#output layer
model.add(tf.keras.layers.Dense(10,activation=tf.nn.softmax))

training_epochs = 30
model.compile(loss='categorical_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])

training_model = model.fit(X_train, y_train,
                     epochs = training_epochs,
                     verbose = 2,
                     validation_data=(X_val, y_val))

predictions=model.predict(X_test)
print(predictions)

print(np.argmax(predictions[11]))

#pd.dataframe converts output to a dataframe
pred = pd.DataFrame(model.predict(X_test))
pred = pd.DataFrame(pred.idxmax(axis = 1))
pred.index.name = 'ImageId'
pred = pred.rename(columns = {0: 'Label'}).reset_index()
pred['ImageId'] = pred['ImageId'] + 1

pred.head(20)

pred.to_csv('/content/drive/My Drive/final_submission.csv', index = False)