Getting Assertion Error

[Arpitrf]

Traceback (most recent call last):
File "training.py", line 203, in
validation_data=(process_features(data.validation, data.metadata), data.validation_labels))
File "/Users/apple/Desktop/Kaggle/Taxi Trajectory/vir/lib/python3.6/site-packages/keras/engine/training.py", line 950, in fit
batch_size=batch_size)
File "/Users/apple/Desktop/Kaggle/Taxi Trajectory/vir/lib/python3.6/site-packages/keras/engine/training.py", line 671, in _standardize_user_data
self._set_inputs(x)
File "/Users/apple/Desktop/Kaggle/Taxi Trajectory/vir/lib/python3.6/site-packages/keras/engine/training.py", line 575, in _set_inputs
assert len(inputs) == 1
AssertionError

--Getting the above error. Any help is appreciated.

[karafede]

here my solution that worked pretty well... using API KERAS functional model:

def create_model(metadata, clusters):
"""
Creates all the layers for our neural network model.
"""

# metadata = data.metadata
# Arbitrary dimension for all embeddings
embedding_dim = 10

# Quarter hour of the day embedding
i_embed_quarter_hour = Input(shape=(1,))
embed_quarter_hour = Embedding(metadata['n_quarter_hours'], embedding_dim, input_length=1)(i_embed_quarter_hour)
quarter_hour_vec = Flatten()(embed_quarter_hour)

# Day of the week embedding
i_embed_day_of_week = Input(shape=(1,))
embed_day_of_week = Embedding(metadata['n_days_per_week'], embedding_dim, input_length=1)(i_embed_day_of_week)
day_of_week_vec = Flatten()(embed_day_of_week)

# Week of the year embedding
i_embed_week_of_year = Input(shape=(1,))
embed_week_of_year=Embedding(metadata['n_weeks_per_year'], embedding_dim, input_length=1)(i_embed_week_of_year)
week_of_year_vec = Flatten()(embed_week_of_year)

# Client ID embedding
i_embed_client_ids = Input(shape=(1,))
embed_client_ids=Embedding(metadata['n_client_ids'], embedding_dim, input_length=1)(i_embed_client_ids)
client_ids_vec = Flatten()(embed_client_ids)

# Taxi ID embedding
i_embed_taxi_ids = Input(shape=(1,))
embed_taxi_ids=Embedding(metadata['n_taxi_ids'], embedding_dim, input_length=1)(i_embed_taxi_ids)
taxi_ids_vec = Flatten()(embed_taxi_ids)

# Taxi stand ID embedding
i_embed_stand_ids = Input(shape=(1,))
embed_stand_ids=Embedding(metadata['n_stand_ids'], embedding_dim, input_length=1)(i_embed_stand_ids)
stand_ids_vec = Flatten()(embed_stand_ids)

# GPS coordinates (5 first lat/long and 5 latest lat/long, therefore 20 values)
i_coords = Input((20,))
coords = Dense(1)(i_coords)


# Merge all the inputs into a single input layer

merged_layer = Concatenate()([
    quarter_hour_vec,
    day_of_week_vec,
    week_of_year_vec,
    client_ids_vec,
    taxi_ids_vec,
    stand_ids_vec,
    coords])

# Final activation layer: calculate the destination as the weighted mean of cluster coordinates
cast_clusters = K.cast_to_floatx(clusters)
def destination(probabilities):
    return tf.matmul(probabilities, cast_clusters)

i_output_layer = Dense(500, activation='relu')(merged_layer)
output_layer = Dense(len(clusters), activation='softmax')(i_output_layer)
output_layer_final = Activation(destination)(output_layer)

model = Model(inputs=[i_embed_quarter_hour,
                      i_embed_day_of_week,
                      i_embed_week_of_year,
                      i_embed_client_ids,
                      i_embed_taxi_ids,
                      i_embed_stand_ids,
                      i_coords], outputs=output_layer_final)

# Compile the model
optimizer = SGD(lr=0.01, momentum=0.9, clipvalue=1.)  # Use `clipvalue` to prevent exploding gradients
model.compile(loss=tf_haversine, optimizer=optimizer)

return model