Implement FNN training agent

[robilkot]

Trainer agent is now working.

Capabilities

Specification of FNN structure in kb

Specification of preproccessed csv-datasets in kb

Specification of training parameters in kb

Given above parameters, agent is able to build tensorflow.keras FNN according to provided structure and train it on given dataset. Input and output data shapes are identified automatically. Trained model is then stored in filesystem, as well as linked to FNN structure in database, allowing for future interpretation

Example output during model training:

FNN structure per-layer is shown, as well as training state on each epoch

MNIST Fashion dataset is used as example.

Using training agent

1. Formalize desired FNN structure by providing layers with their sizes and activation functions in ordered way
2. Provide dataset: specify filepath and labels column
3. Provide desired training parameters: include learning rate, batch size and number of epochs
4. Create edge from 'question_initiated' node to action node
5. Wait for appearence of link containing path to file with trained keras model

Trained models are stored at kb\fnn_processing_module\trained_models

[robilkot]

Moved docker changes to #35