A Review of Network Inference Techniques for Neural Activation Time Series

The scripts follow the order below. If two scripts have the same number, they can be run simultaneously.
The code was developed using python 2.7, in an ubuntu mate environment and tested in a laptop with Intel i7 CPU@2.4 GHz, 8 Gb RAM

Python packages used: pandas, sklearn, numpy, tensorflow, pickle, pyhawkes
OASIS for spike inference from https://github.com/j-friedrich/OASIS
RCNN is accompanied with folders:
-conutils from https://github.com/spoonsso/TFconnect
-tfomics and model_zoo from https://github.com/spoonsso/tfomics

The folder structure of the project is :
Code-> The content of this folder
Data-> Download the "small" dataset from https://www.kaggle.com/c/connectomics/data and extract it here. Each network has three .txt files: the neuron's activations, positions and ground truth connectivity
Results-> Will be filled with estimated connectivity matrices, evaluation metrics and time logs

1 utils contains functions (either costum or found online, with appropriate reference) for data preprocessing and evaluation of the algorithms

2 preprocess runs the discretization algorithms and produces discritized versions of the activation series

3 run_model_free runs correlation-based approaches

3 run_hawkes runs hawkes process model using on pyhawkes

3 run_rcnn runs residual convolutional neural network with the aid of conutiles, tfomics and model_zoo

3 run_influence runs the model that is based on influence estimation in social networks

4 evaluate uses the predicted connectivity matrices stored in results and the ground truth to calculate evaluation metrics

5 plot_connectivity plots the predicted connectivity matrices and the ground truth