tfmodel | Canned estimators and pre-trained models | Machine Learning library

You are using the dataset int fit instead of train_data. I assume you are using a DataFrame called X_train and y_train and I mimicked the same with numpy and it works now. See below.

Firestore`s queries run asyncronously, not one after another. So the second query may start earlier than the first is completed.

If you want to run them one by one you need to put 2nd query into 1st.

Try this:

It was because of silly compiler optimization. I added the proguard settings, and everything works fine.

https://developer.android.com/studio/build/shrink-code#keep-code

.pro file

This is wrong API usage.

Changing typed_input_tensor to typed_tensor and typed_output_tensor to typed_tensor resolved the issue for me.

For anyone else having the same issue,

I think you are talking about two things:

1. the model itself. I assume your model can fit in your memory. Otherwise you could not run any prediction.
2. the data. If data is the problem, you should make a data generator with python so that not all the data exist in the memory at the same time. You should generate each example (x) or each batch of examples and feed them into the model to get prediction. The result could be serialized to disk when necessary if your memory cannot hold all results.

More concretely, something like this:

Okay I finally solved it that way:

Is there anything I'm missing in regard to this different representation logic which makes the practice confusing?

In fact, you made a mistake about the input shapes of static_rnn and dynamic_rnn. The input shape of static_rnn is [timesteps,batch_size, features](link),which is a list of 2D tensors of shape [batch_size, features]. But The input shape of dynamic_rnn is either [timesteps,batch_size, features] or [batch_size,timesteps, features] depending on time_major is True or False(link).

Could the solution be attained to switching to dynamic_rnn?

The key is not that you use static_rnn or dynamic_rnn, but that your data shape matches the required shape. The general format of placeholder is like your code is [None, N_TIMESTEPS_X, N_FEATURES]. It's also convenient for you to use dataset API. You can use transpose()(link) instead of reshape().transpose() will permute the dimensions of an array and won't messes up with the data.

So your code needs to be modified.

The basic iterative method (BIM) applies the fast gradient sign method (FGSM) multiple times (100 times with the parameters that you have specified). Each step of the BIM applies the FGSM on the outcome of the previous step of the BIM. Therefore, your model object needs to have a method fprop that returns the output of the model for any input tensor passed as an argument. The current class you have implemented always returns the output of the model on the same placeholder self.x. You will have to use scopes to define a fprop method that can take an arbitrary tensor x and return the output of the model on that input. You can find an example of a simple model implementation ModelBasicCNN that does that in the tutorials folder: https://github.com/tensorflow/cleverhans/blob/master/cleverhans_tutorials/tutorial_models.py

You need to set the session to the keras TF backend