Gradient Tape

Gradient_tape.py

inputs = Input(shape=(7))
fc = Dense(units=72)(inputs)
fc = Dense(units=1, activation='sigmoid')(fc)
model = Model(inputs, fc)
#generating data
X = np.random.rand(1024, 7)
#generating labels
y = np.conacatenate(np.zeros((512)), np.ones((512)))
optimizer = Adam()

#Train the model
for epoch in range(epochs):
  with tf.GradientTape() as tape:
    #getting predictions
    pred = model(X)
    #computing loss
    loss = your_loss_fun(y,pred)
  #calculate gardients
  gradients = tape.gradient(loss, model.trainable_variables)
  #Apply gradients and optimize
  optimizer.apply_gradients(zip(gradients, model.trainable_variables))