Arkajyoti Mukherjee arkajyotiMukherjee

💭

githubProfile.welcome();

contributor to @litmuschaos | web developer | definitely not batman (wouldn't tell you if I were)

arkajyotiMukherjee / custom_call_train_keras_article.py

Created June 28, 2019 09:59

	# Compile
	model.compile(optimizer='adam',
	loss='categorical_crossentropy',
	metrics=['accuracy'])
	# Train
	model.fit(train_data, train_labels, epochs=10, callbacks=callbacks,
	validation_data=(test_data, test_labels))
	# check the prediction for the 500th test example
	check_pred(500)

arkajyotiMukherjee / custom_callback_keras_article.py

Created June 28, 2019 09:50

	# Writing a learning rate reducer function, takes epoch as parameter
	def reducer(epoch):
	# increase epoch by 1 as epochs are counted from 0
	epoch = epoch + 1
	if epoch<5:
	return 0.001
	else:
	# the return of the scheduler must be a float
	return 0.001 * np.exp(0.1 * (5 - epoch))

arkajyotiMukherjee / model_sub_train_keras_article.py

Created June 28, 2019 09:34

	# Create the model instance
	model = MyModel(num_classes=10)

	# Compile
	model.compile(optimizer=tf.train.RMSPropOptimizer(0.001),
	loss='categorical_crossentropy',
	metrics=['accuracy'])
	# Trains for 5 epochs.
	model.fit(train_data, train_labels, batch_size=32, epochs=5)
	model.evaluate(test_data, test_labels)

arkajyotiMukherjee / model_sub_keras_article.py

Created June 28, 2019 09:29

	class MyModel(tf.keras.Model):
	def __init__(self, num_classes):
	super(MyModel, self).__init__(name='my_model')
	self.num_classes = num_classes
	# Defining the layers
	self.conv_1 = layers.Conv2D(filters=32, kernel_size=(3,3), activation='relu', input_shape=(28,28,1))
	self.maxpool_1 = layers.MaxPooling2D(pool_size=(2,2))
	self.conv_2 = layers.Conv2D(filters=64, kernel_size=(3,3), activation='relu')
	self.maxpool_2 = layers.MaxPooling2D(pool_size=(2,2))
	self.conv_3 = layers.Conv2D(filters=128, kernel_size=(3,3), activation='relu')

arkajyotiMukherjee / pred_for_1st_keras_article.py

Last active June 28, 2019 09:36

	def check_pred(index):
	# Testing the prediction for the 1st number
	plt.figure()
	plt.imshow(np.squeeze(test_data[index]), cmap='gray')
	# We are using np.argmax since the labels are in categorical form
	plt.title('Number: {}'.format(np.argmax(results[index])), color='magenta')
	plt.axis('off')
	plt.show()

	# check the prediction for the 0th test example

arkajyotiMukherjee / func_api_train_keras_article.py

Last active June 28, 2019 09:37

	# Compile
	model.compile(optimizer=tf.train.AdamOptimizer(learning_rate=0.001),
	loss=tf.keras.losses.categorical_crossentropy,
	metrics=['accuracy'])
	# Train
	model.fit(train_data, train_labels, batch_size=32, epochs=5)
	# Evaluate
	model.evaluate(test_data, test_labels)
	# check the prediction for the 1st test example
	check_pred(1)

arkajyotiMukherjee / func_api_keras_article.py

Created June 28, 2019 08:43

	# Make an input tensor
	inputs = tf.keras.Input(shape=(28,28,1))

	# A layer instance is callable on a tensor, and returns a tensor.
	x = layers.Conv2D(filters=32, kernel_size=(3,3), strides=2, activation='relu')(inputs)
	# We have to provide the input for each layer manually
	# This is how we can make arbitrary connections
	x = layers.Conv2D(filters=64, kernel_size=(3,3),strides= 2, activation='relu')(x)
	# Notice how we are using the same variable x, as python performs the operation first then the assignment
	x = layers.Conv2D(filters=64, kernel_size=(3,3),strides=2, padding='same', activation='relu')(x)

arkajyotiMukherjee / pred_eval_keras_article.py

Created June 28, 2019 07:02

	# Make Predictions
	results = model.predict(test_data)
	print('Prediction shape: ', results.shape)

	# Evaluate the model
	print('Model evaluation: ')
	model.evaluate(test_data, test_labels, batch_size=32)

arkajyotiMukherjee / model_fit_keras_article.py

Created June 28, 2019 06:51

	model.fit(train_data,
	train_labels,
	epochs=5,
	batch_size=32,
	validation_data=(test_data, test_labels))

arkajyotiMukherjee / compile_keras_article.py

Created June 28, 2019 06:39

	# We can configure the model training in a variety of ways
	model.compile(optimizer=tf.train.AdamOptimizer(0.01),
	loss='categorical_crossentropy',
	metrics=['accuracy'])

	# Configure a model for categorical classification.
	model.compile(optimizer=tf.train.RMSPropOptimizer(0.01),
	loss=tf.keras.losses.categorical_crossentropy,
	metrics=[tf.keras.metrics.categorical_accuracy])