Keras dev example

[Onoyiza]

Hi @pgleeson.

I have done the following:

• Updated the keras_to_mdf function to include the activation within the dense node calling on the "add_activation" function
• After various trials on why the output was not the same for the IRIS dataset I discovered that to handle numeric instability that occurs due to large exponential, Keras first centers the values in the numpy array or tensor around 0 by subtracting by the maximum value before getting the exponential and dividing by its sum. I tried to subtract the max value from the array in the script with the keras_to_mdf function but was unsuccessful because the expression could not be evaluated due to the addition of "max" in the string(expression). The code where I created the expression to be evaluated for subtracting the max value from each value in the array is a comment in the script.
• I downloaded and ran the notebook(and script) on the 3-D convolutional model on the CT-scan dataset and I have saved the model in h5 format to use to test when I am done creating the respective nodes for the layers (Conv, MaxPool, BatchNorm, GloabalAveragePool, Dropout) in the model. I wanted to successfully get the softmax function to work well before I start creating the nodes for this task but I want unsuccessful with it. I just started working on this since I noticed uses the sigmoid function for classification.

[pgleeson]

Thanks @Onoyiza. Merging in current changes and will test locally also.