[BUG] about ydata-synthetic

[fabclmnt]

Hi @Sorooshi this is not a bug.

As you can see per the error returned, the GAN architecture is expecting an object with the following structure: (batch_size, 24, 10).

ValueError: Input 0 is incompatible with layer model: expected shape=(None, 24, 10), found shape=(69, 1, 10)

You're passing an object with the structure (69,1,10). The processing provided is not generalizable, meaning it was build for the example dataset. It's expected that the preprocessing is adapted to adapt the new data into the structure expected by the network.

In the future we will provide more generalizable processing methods. For the moment we only ensure that the example dataset will work.

[Sorooshi]

@fabclmnt Correct me if I'm mistaken: this error arises not because of the preprocessing but because of the TGAN code, that is when you train the autoencoder!

It has nothing to do with preprocessing

[fabclmnt]

The error is still due to the preprocessing - meaning the architecture is expecting a structure on the vector side if you are providing a structure that is not compatible with what we are expecting at the TGAN level, the problem is from the preprocessing.

The architecture was developed under certain assumptions - (batchs_size, 24, 20). Unfortunately, this is the limitation we have for the moment.