Access Python Library importing Keras

Hi all,

I am triyng to call python scripts i allready wrote in matlab.

In specific i am trying call i python function which imports keras models i created and uses them to predict the input data like this:

 from keras.models import load_model
 from euclid_dist import euclid_dist
 import numpy as np

 class nn_handler:

    def __init__(self, model_location='current_checkpoint_lstm_2'):
        self.model = load_model(model_location,custom_objects={'euclid_dist':euclid_dist})

    def prediction(self,X):
        pred=self.model.predict(X)
        return pred.reshape(pred.shape[0],9,3)

However i noticed that matlab is (amoungst others) not accepting python scripts importing keras.

For exmple if i write two scripts:

test.py:

 import numpy

 def foo():
    return 'bar'

and test2.py: (i only change the import and are not even using it)

 import keras

 def foo():
    return 'bar'

And call them in matlab, setting the version to my anaconda enviroment:

 pyversion('C:\Users\lukas.scheunemann\AppData\Local\Continuum\miniconda3\envs\nnenv\pythonw.exe')

 py.test.foo()
 py.test2.foo()

The first one returns 'bar'. But the secound an error 'Undefined variable "py" or class "py.test2.foo".

Did i make a misstake installing keras so it can not be used?

Or is only a limited selection off python libraries supported unlike the official limitations , because it is not desired that people use python instead off aquivilent matlab packages. E.g. i am supposed to buy the deep learning toolbox.

[ARUN]

I am facing the same probelm, you got this resolved? I am trying to call the python script which contains the keras model imported through load_model as yours. I am getting the same undefined function error.

Answer 1

[Lukas Scheunemann]

I was able to solve the task by writting the arguments into a .mat file and then start the python script via the systems() command. Which in turn loads from the .mat and the model, then makes a prediction and returns the result again in a .mat file, which is loaded in matlab again.

It is slow and a should have given a warning to everybody valueing their sanity bevor make them even read this ... but it works.

[LEONARDO NUTI]

Hey i'm facing same issue can you elaborate on how you saved arguments in .mat file? I've xgboost and nn models written in Python that I need to run thorugh MATLAB