Data cardinality is ambiguous: Keras model prediction with multiple time series and vector inputs

Fakhreddin Amirhosseini 0
Viewed 31

I have a model that takes in two inputs, one is a sequence of vectors and the other one is a simple vector, and my output is also a simple vector. E.g. x1.shape: (50000,50,2) x2.shape: (50000,2) y.shape: (50000,2)

I have successfully trained and evaluated my model using mode.fit and model.evaluate with no errors as follows:

model.fit(
    x=[x1[:-validationLength], x2[:-validationLength]],
    y=y[:-validationLength],
    epochs=25,
    batch_size=256,
    validation_split=0.2,
)

model.evaluate(x=[x1[-validationLength:], x2[-validationLength:]],
    y=y[-validationLength:],
    batch_size=2)

2147/2147 [==============================] - 18s 9ms/step - loss: 0.0560 - accuracy: 0.8498

But when I try to use my model:

output = model.predict([x1[1000], x2[1000]])

the following error pops up:

ValueError: Data cardinality is ambiguous:
  x sizes: 50, 2
Make sure all arrays contain the same number of samples.

IDK how this possible!!!!!

Here is a diagram of the model: enter image description here

any suggestions will be deeply appreciated!

lstm keras tensorflow python
1 Answers

Maybe you are missing the batch dimension and your first dimensions are being interpreted as the batch dimension, try:

output = model.predict([x1[None, 1000], x2[None, 1000]])
AloneTogether0
Related