thierryherrmann

np.random.seed(2); tf.random.set_seed(5)

def make_model():
    # this constructs a keras Model. We use the functional API and add a custom
    # layer for demo purposes but a model of any complexity can be used here
    from tensorflow.keras import layers
    
    class CustomLayer(keras.layers.Layer):
        def __init__(self, **kwargs):
            super().__init__(**kwargs)

thierryherrmann

for epoch in range(epochs):
    print("\nStart of epoch %d" % (epoch,))
    start_time = time.time()

    # Iterate over the batches of the dataset.
    for step, (x_batch_train, y_batch_train) in enumerate(train_dataset):
        with tf.GradientTape() as tape:
            logits = model(x_batch_train, training=True)
            loss_value = loss_fn(y_batch_train, logits)
        grads = tape.gradient(loss_value, model.trainable_weights)

thierryherrmann

# This was tested from tf-nightly 2.1.0.dev20191111 (Linux Ubuntu 18.04)
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
from scipy import stats
import os

import tensorflow as tf
from tensorflow import keras

thierryherrmann

# https://github.com/tensorflow/tensorflow/issues/33150
import tensorflow as tf

class Net(tf.keras.Model):
    def __init__(self):
        super(Net, self).__init__()
        self.l1 = tf.keras.layers.Dense(5)
    def call(self, x):
        return self.l1(x)