process_data_graph.py

import os
import torch
import pandas as pd
import numpy as np
from getData import getData
from pca_process import transformation_data
import multiprocessing


device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

def main(methods, transform_methods, database, db_name):
    for method in methods:
        for transform_method in transform_methods:
            data = getData()

            print('Procesando metodo: ', method)
            if method == '':
                method = 'seq_to_seq'
                name = 'DB: PIP y features ' + method
                g = data.seq_to_seq(database)
                if transform_method == 'pca':
                    transform_instance = transformation_data()
                    data_transform_pca, pca_instance = transform_instance.apply_pca_linear(g.ndata['feat'])
                    data_transform_pca = np.round(data_transform_pca, 3)
                    g.ndata['feat'] = torch.Tensor(data_transform_pca)
                elif transform_method == 'kernel_pca':
                    transform_instance = transformation_data()
                    data_transform_pca, pca_instance = transform_instance.apply_kernel_pca(g.ndata['feat'])
                    data_transform_pca = np.round(data_transform_pca, 3)
                    g.ndata['feat'] = torch.Tensor(data_transform_pca)
            else:
                name = 'DB: PIP y features ' + method
                g = data.encoding(method, database)
                if transform_method == 'pca':
                    transform_instance = transformation_data()
                    data_transform_pca, pca_instance = transform_instance.apply_pca_linear(g.ndata['feat'])
                    data_transform_pca = np.round(data_transform_pca, 3)
                    g.ndata['feat'] = torch.Tensor(data_transform_pca)
                elif transform_method == 'kernel_pca':
                    transform_instance = transformation_data()
                    data_transform_pca, pca_instance = transform_instance.apply_kernel_pca(g.ndata['feat'])
                    data_transform_pca = np.round(data_transform_pca, 3)
                    g.ndata['feat'] = torch.Tensor(data_transform_pca)

            # CARPETA DONDE QUEDA CADA SET DE DATOS
            ruta = db_name + '_data_modelos/'
            if not os.path.exists(ruta):
                os.mkdir(ruta)
            ruta = db_name + '_data_modelos/' + transform_method
            if not os.path.exists(ruta):
                os.mkdir(ruta)
            ruta = db_name + '_data_modelos/' + transform_method + '/' + method
            if not os.path.exists(ruta):
                os.mkdir(ruta)
            
            nodes_features = pd.DataFrame(np.round(g.ndata['feat'].numpy(), 3))
            nodes_features.to_csv(db_name + '_data_modelos/' + transform_method + '/' + method + '/nodes_features_.csv', index=False)
            
            if not os.path.isfile(db_name + '_data_modelos/edge_index.csv'):
                print('Se crea edge_index')
                edge_index = pd.DataFrame()
                src, dst = g.edges()
                train_edges = torch.cat([src.unsqueeze(0), dst.unsqueeze(0)], dim=0)
                edge_index['edge_1'] = train_edges.tolist()[0]
                edge_index['edge_2'] = train_edges.tolist()[1]
                edge_index.to_csv(db_name + '_data_modelos/edge_index.csv', index=False)
            
            if not os.path.isfile(db_name + '_data_modelos/weights.csv'):
                print('Se crea edge_weights')
                edge_weights = pd.DataFrame()
                edge_weights['weights'] = g.edata['weight'].tolist()
                edge_weights.to_csv(db_name + '_data_modelos/weights.csv', index=False)

            if not os.path.isfile(db_name + '_data_modelos/labels.csv'):
                print('Se crea edge_labels')
                edge_labels = pd.DataFrame()
                edge_labels['labels'] = g.edata['label'].tolist()
                edge_labels.to_csv(db_name + '_data_modelos/labels.csv', index=False)


if __name__ == "__main__":
    ruta = 'Protein-Protein/encoded_sequences'
    methods = [] # methods = [''] vacio es seq_to_seq
    
    for elemento in os.listdir(ruta):
        if os.path.isdir(os.path.join(ruta, elemento)):
            methods.append(elemento)

    # transform_method = ['no_transform', 'pca', 'kernel_pca']
    transform_methods = ['no_transform'] # sin pca ni kernel_pca

    databases = ['kd_pdbbind_database.csv', 'pip_database.csv', 'proximate_dg.csv', 'proximate_kd.csv',
                 'proximate_kon.csv', 'skempi_affinity.csv', 'skempi_koff.csv', 'skempi_kon.csv'] # no se incluye pip

    # methods = ['bepler']
    # databases = ['kd_pdbbind_database.csv']    
    processes = []
    for database in databases:
        db_name = database.split('.')[0]
        process = multiprocessing.Process(target=main, args=(methods, transform_methods, database, db_name))
        processes.append(process)
        process.start()

    for process in processes:
        process.join()