Hyperparameter Optimization for Keras model with large dataset about talos HOT 7 CLOSED

bump.

I'm fairly new to DL and Keras (but not new to other AI and ML) and I gotta say that it is amazing that most tools, examples, and academic papers are all built around these teeny tiny "toy" datasets like MNIST and CIFAR. Who uses 32x32 pixel images in a real application?

Please add my vote for adding Keras Sequence support in Talos so it can be used in real applications.

from talos.

@NTNguyen13 Indeed it is something being worked on. Just curious: how large is your dataset exactly?

from talos.

Hi, @x94carbone my dataset has 11000 images, each with 30-40 KB

from talos.

Hmm ok. Does your model work with Keras' fit on its own? Without using fit_generator?

from talos.

I tried it with small random data and it works, it didn't work in the real case because of the large dataset

from talos.

So this likely means that there's nothing wrong with Talos at the moment, we've just gotta implement a feature to get fit_generator to work. Feel free to let me know if I'm missing something here.

from talos.

Looks like all clear so closing here.

from talos.