简体   繁体   中英

gradient using tf.GradientTape() wrt inputs is None (Tensorflow 2.4)

This is my model. Im using Tensorflow 2.4.1.

model = tf.keras.Sequential([
    tf.keras.layers.Embedding(input_dim=1000,
                              output_dim=64,
                              name='embedding',
                              mask_zero=True),
    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(32)),
    tf.keras.layers.Dense(16, activation='relu'),
    tf.keras.layers.Dense(4, name='logits')
])

metrics = [tf.keras.metrics.SparseCategoricalAccuracy()]

# compile the model
model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=metrics)

when I run the following code, I get None as gradient wrt input.

def compute_gradients(t, target_class_idx):
    with tf.GradientTape() as tape:
        tape.watch(t)
        logits = model(t)
        probs = tf.nn.softmax(logits, axis=-1)[:, target_class_idx]
    grads = tape.gradient(probs, t)
    return grads

Here is a sample input, and the call

sample_tensor = tf.random.uniform(shape=(1, 50))

path_gradients = compute_gradients(
    t=sample_tensor,
    target_class_idx=0)

print(path_gradients)

None

What am I doing wrong ?

Thanks

The Embedding layer in TensorFlow is not differentiable. Source: https://github.com/keras-team/keras/issues/12270

The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.

 
粤ICP备18138465号  © 2020-2024 STACKOOM.COM