multi output forcast shape prob. with keras.preprocessing.timeseries_dataset_from_array

Question

The model was fuctioinnal with only one output, now I try to use multi output forcasting one timestep, but got this error:

Epoch 1/10
2025-11-04 16:20:09.341769: W tensorflow/core/framework/op_kernel.cc:1844] INVALID_ARGUMENT: required broadcastable shapes
InvalidArgumentError: Graph execution error:

Detected at node compile_loss/mse/sub defined at (most recent call last):
  File "/databricks/python_shell/scripts/db_ipykernel_launcher.py", line 52, in <module>

My input data are build this way from a pandas df:

train_data = features.loc[0 : train_split - 1]
val_data = features.loc[train_split:]

train_data.shape
(500000, 10)

x_train = train_data.iloc[:, :6].values
y_train1 = train_data.iloc[:,6].values
y_train2 = train_data.iloc[:,7].values
y_train3 = train_data.iloc[:,8].values
y_train4 = train_data.iloc[:,9].values
combined_targets_train = np.stack([y_train1, y_train2, y_train3, y_train4], axis=-1)

dataset_train = keras.preprocessing.timeseries_dataset_from_array(
    x_train,
    combined_targets_train,
    sequence_length=sequence_length,
    sampling_rate=step,
    batch_size=batch_size,
    shuffle=True,
)

for batch in dataset_train.take(1):
    inputs, targets = batch

print("Input shape:", inputs.numpy().shape)
print("Target shape:", targets.numpy().shape)
Input shape: (16, 120, 6)
Target shape: (16, 4)

In the same way, validation set was build and got this shape Input shape: (16, 120, 6) Target shape: (16, 4)

The model simplify version:

inputs_seq = keras.layers.Input(shape=(120, 6), name='meteo')
lstm_out = LSTM(10, return_sequences=True)(inputs_seq )
...

output_surfrun = Dense(1, activation='linear', name='pred_surf_runoff')(lstm_out )
output_subsurf = Dense(1, activation='linear', name='pred_sub_surf_runoff')(lstm_out )
output_evap = Dense(1, activation='linear', name='pred_evapo')(lstm_out )
output_een = Dense(1, activation='linear', name='pred_een')(lstm_out )

model = keras.Model(inputs=inputs_seq, outputs=[output_surfrun, output_subsurfacerun, output_evap, output_een], name='model_multi_heads')

model.compile(optimizer=keras.optimizers.Adam(learning_rate=learning_rate), loss="mse")

history = model.fit(
    dataset_train,
    epochs=10,
    validation_data=dataset_val
)

I make many shape validation but don't find what the problem.

Vy Do · Accepted Answer · 2025-11-06 02:20:25Z

1

Issue 1: Time dimension mismatch: your code return_sequences=True , therefore LSTM return all of ouput series in format of (batch, timesteps, units) , more specific (B, 120, 10) . After go through Dense(1) , return (B, 120, 1) . While target in format (B, 4).

Solution:

Remove return_sequences=True or
Only get latest step.

Issue 2

Multiple inputs, but only one output targeted to format (B, 4) .

Solution

Convert (B, 4) to 4 tensors (B, 1) or,
Using one Dense(4) only.

edited Nov 6 at 2:20

answered Nov 6 at 2:15

Vy Do

1

Sign up to request clarification or add additional context in comments.

2 Comments

Jonathan Roy Nov 11 at 16:05

Issue 2, Convert (B,4) to 4 tensors (B,1), this is not what I do with this line?? :combined_targets_train = np.stack([y_train1, y_train2, y_train3, y_train4], axis=-1)

Jonathan Roy Nov 11 at 20:01

finaly found how, turn all my label into new df like this: combined_targets = pd.DataFrame(dtset.iloc[:,6:]) in lieu of combined_targets_train = np.stack([y_train1, y_train2, y_train3, y_train4], axis=-1)

Collectives™ on Stack Overflow

multi output forcast shape prob. with keras.preprocessing.timeseries_dataset_from_array

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related