Model_training&Heatmap.py

# -*- coding: utf-8 -*-
"""Project Training.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1Kr15oI3Jz958EB30I3We9V278wZnGd83
"""

#!pip install tensorflow-gpu
import sklearn as sk
from sklearn.metrics import confusion_matrix

import seaborn as sns

import sys
import numpy

import matplotlib.pyplot as plt

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.models import Sequential # For constructing model
from tensorflow.keras.layers import Dense, Dropout, Flatten # Layer cores
from tensorflow.keras.layers import Conv2D, MaxPooling2D # CNN layers
from tensorflow.keras.utils import to_categorical # Extra utilities

import pickle
from sklearn.model_selection import train_test_split

import os

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

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

# GPU test code
device_name = tf.test.gpu_device_name()
if device_name != '/device:GPU:0':
  raise SystemError('GPU device not found')
print('Found GPU at: {}'.format(device_name))

def loadData(fileName,size=0.2):
    with open('/content/drive/My Drive/X_Y_Data.pickle', 'rb') as f:
        X, Y = pickle.load(f)
    
    X=X.reshape(-1,45,45,1)
    X_train, X_test, y_train, y_test = train_test_split(X,Y,test_size = size)
    return X_train, X_test, y_train, y_test

def createModel(size):

    model = Sequential()

    # Images are 48 by 48
    model.add(Conv2D(32, (3,3), activation='relu', input_shape=size)) #46 by 46
    model.add(MaxPooling2D())
    model.add(Conv2D(64, (3,3), activation='relu')) #44 by 44
    model.add(MaxPooling2D())
    model.add(Conv2D(128, (3,3), activation='relu')) #44 by 44
    model.add(MaxPooling2D())
    model.add(Dropout(rate=0.15))
    model.add(Flatten()) #1964 by 1
    model.add(Dense(500, activation='relu')) #500 by 1
    model.add(Dropout(0.2))
    model.add(Dense(250, activation='relu')) #250 by 1
    model.add(Dropout(0.2))
    model.add(Dense(125, activation='relu')) #120 by 1
    model.add(Dropout(0.2))
    model.add(Dense(66, activation='softmax')) # 66 by 1 (only english, digits, and symbols)
    
    model.compile(optimizer='adam',
                loss='sparse_categorical_crossentropy',
                metrics=['accuracy'])
    
    return model

X_train, X_test, y_train, y_test = loadData('X_Y_Data.pickle')
model = createModel(X_train.shape[1:])

checkpoint_path = "training/cp-{epoch:04d}.ckpt"

checkpoint_dir = os.path.dirname(checkpoint_path)

cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_path, verbose=1, save_weights_only=True, period = 2)

model.save_weights(checkpoint_path.format(epoch=0))

model.fit(X_train, y_train, epochs=16, callbacks=[cp_callback], validation_split = .2)

loss, acc = model.evaluate(X_test, y_test)

# Get Confusion matrix
con_mat = confusion_matrix(y_test, model.predict(X_test).argmax(axis=1))
con_mat = con_mat.astype(float)
# numpy.set_printoptions(threshold=sys.maxsize)
# print(con_mat)

# Normalize confusion matrix row wise
for r in range(66):
  s = float(sum(con_mat[r, :]))
  for c in range(66):
    con_mat[r, c] = float(con_mat[r, c])/s

#print(con_mat)

# Create heat map of confusion matrix
plt.figure(figsize=(18, 14))
sns.heatmap(con_mat, square=True, linewidths=0.01, linecolor='#A9A9A9')