DoctorLoop

def model(data):
    fc1w_prior = dist.Normal(loc=torch.zeros_like(dense.fc1.weight), scale=torch.ones_like(dense.fc1.weight))
    fc1b_prior = dist.Normal(loc=torch.zeros_like(dense.fc1.bias), scale=torch.ones_like(dense.fc1.bias))
    
    fc2w_prior = dist.Normal(loc=torch.zeros_like(dense.fc2.weight), scale=torch.ones_like(dense.fc2.weight))
    fc2b_prior = dist.Normal(loc=torch.zeros_like(dense.fc2.bias), scale=torch.ones_like(dense.fc2.bias))
    
    fc3w_prior = dist.Normal(loc=torch.zeros_like(dense.fc3.weight), scale=torch.ones_like(dense.fc3.weight))
    fc3b_prior = dist.Normal(loc=torch.zeros_like(dense.fc3.bias), scale=torch.ones_like(dense.fc3.bias))
    

DoctorLoop

for epoch in range(800): 

    running_loss = []
    for i, batch in enumerate(trainloader):
        
        inputs = batch["features"]
        labels = batch["outcomes"]

        optimizer.zero_grad()

DoctorLoop

class DenseModel(nn.Module):
    def __init__(self, num_in):

        super(DenseModel, self).__init__()

        self.fc1 = nn.Linear(num_in, 16)
        self.fc2 = nn.Linear(16, 6)
        self.fc3 = nn.Linear(6, 3)

    def forward(self, x):

DoctorLoop

tf.keras.backend.clear_session()

dataset_size = len(train_)
kl_divergence_function = (lambda q, p, _: dist.kl_divergence(q, p) / tf.cast(dataset_size, dtype=tf.float32))

model_tfp = tf.keras.Sequential([
    tf.keras.Input(shape=(2,),name="basket"),
    tfp.layers.DenseFlipout(16, kernel_divergence_fn=kl_divergence_function, activation=tf.nn.relu, name="dense_tfp_1"),
    tfp.layers.DenseFlipout(6, kernel_divergence_fn=kl_divergence_function, activation=tf.nn.relu, name="dense_tfp_2"),
    tfp.layers.DenseFlipout(3, kernel_divergence_fn=kl_divergence_function, activation=tf.nn.softmax, name="out_tfp_pred"),

DoctorLoop

tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir="./logs/egg_times", histogram_freq=1)

history = model.fit(train_, 
                    target_,
                    epochs=800, 
                    verbose=0, 
                    use_multiprocessing=True, 
                    callbacks=[tensorboard_callback],
                    validation_split=0.1,
                    validation_freq=20)

DoctorLoop

tf.keras.backend.clear_session()

model = tf.keras.Sequential([
    tf.keras.Input(shape=(2,),name="basket"),
    tf.keras.layers.Dense(16, activation="relu", name="dense_1"),
    tf.keras.layers.Dense(6, activation="relu", name="dense_2"),
    tf.keras.layers.Dense(3, activation="softmax", name="out_pred"),
])

learning_rate = 1.0e-4

DoctorLoop

violin_parts = plt.violinplot(underdone['time'], positions=[0], showmeans=True)
# over-ride default blue color, you can't use 'color' parameter as can on histograms etc
for pc in violin_parts['bodies']: # used to
    pc.set_facecolor('red')
    violin_parts['cbars'].set_edgecolor('red')
    violin_parts['cmaxes'].set_edgecolor('red')
    violin_parts['cmins'].set_edgecolor('red')
    violin_parts['cmeans'].set_edgecolor('red')
    plt.violinplot(softboiled['time'], positions=[0.5], showmeans=True)
    plt.violinplot(hardboiled['time'], positions=[1], showmeans=True)