ANN-charge-transfer-coupling

A framework for deployment of artificial neural network with tensorflow for predicting charge transfer coupling.

This is the codes for my final year project.

How to run the code

Add training input data file and output data file (in .npy format) into this directory (also change the FEATURE and RESULT name to your input and output data respectively). Modify hyperparameters in config.py and model architechture in model.py. Create a directory to store the data or set the directory in config (relative path). For windows user, the save path in postprocessing.py might need to be modified. Optionally, change the optimizer in config.py. To start training, run main.py.

input feature for fully connected model (Dense)

any shape

input feature for fully connected model (Dense)

must be 4D array. For example, one channel: shape=(1000, 100, 100, 1) or 3 channel: shape=(1000, 100, 100, 3)

Use other schedular (for training rate control)

Define the schedular in schedular.py, then change the line in main.py import as schedular

Customize the output graph

Change the content in postprocessing.py

Dependancies

python, tensorflow, numpy, matplotlib, sklearn

Others

An example of the result is included in the example file