ValueError: An operation has None for gradient.

[KihaRaito]

Hi!

I tried to set Params.max_passage_count to 1 and write a method for getting word embedding that matches the shape of INPUT from passage_ids and question_ids and model.fit, but I encountered the following gradient error.

src/model.py

model.fit(get_batch("dev"), batch_size=Params.batch_size, epochs=Params.num_epochs, verbose=1)

src/data_load.py

class Embedding:
    ...
    def get_word_emb(self, ids):
        return [getattr(self, "_word_emb")[i] for i in ids]

...

def get_batch(mode="train"):
    data, shapes = load_data("../dev.json")
    input_queue = tf.train.slice_input_producer(data, shuffle=False)
    batch = tf.train.batch(input_queue, num_threads=2,
                           batch_size=Params.batch_size, capacity=Params.batch_size * 32, dynamic_pad=True)
    return batch

Error

Traceback (most recent call last):
  File "model.py", line 72, in <module>
    Vnet()
  File "model.py", line 65, in __init__
    model.fit(get_batch("dev"), batch_size=Params.batch_size, epochs=Params.num_epochs, verbose=1)
  File "/anaconda3/lib/python3.7/site-packages/keras/engine/training.py", line 1010, in fit
    self._make_train_function()
  File "/anaconda3/lib/python3.7/site-packages/keras/engine/training.py", line 509, in _make_train_function
    loss=self.total_loss)
  File "/anaconda3/lib/python3.7/site-packages/keras/legacy/interfaces.py", line 91, in wrapper
    return func(*args, **kwargs)
  File "/anaconda3/lib/python3.7/site-packages/keras/optimizers.py", line 475, in get_updates
    grads = self.get_gradients(loss, params)
  File "/anaconda3/lib/python3.7/site-packages/keras/optimizers.py", line 91, in get_gradients
    raise ValueError('An operation has `None` for gradient. '
ValueError: An operation has `None` for gradient. Please make sure that all of your ops have a gradient defined (i.e. are differentiable). Common ops without gradient: K.argmax, K.round, K.eval.

It looks like verify_loss is being implemented, is it?
src/loss_functions.py

def verify_loss(y_true, y_pred):
    batch_losses = K.sum(y_true * K.log(y_pred), axis=-1)
    # batch_losses = K.map_fn(loss_per_batch, (y_true, y_pred), dtype="float32")
    return -K.mean(batch_losses)