DDI_prediction.py

import csv
import numpy as np
import sys
import pandas as pd
import itertools
import math
import time
from sklearn import svm, linear_model, neighbors
from sklearn import tree, ensemble
from sklearn import metrics
from sklearn.naive_bayes import GaussianNB
import networkx as nx
import random
import numbers
from sklearn.model_selection import StratifiedKFold
from sklearn.neighbors import KNeighborsClassifier
from src import ml

import findspark
findspark.init("/home/spark-2.3.1-bin-hadoop2.7/")

from pyspark import SparkConf, SparkContext

if False: 
    sc.stop()

config = SparkConf()
config.setMaster("local[8]")
config.set("spark.executor.memory", "56g")
config.set('spark.driver.memory', '30g')
config.set("spark.memory.offHeap.enabled",True)
config.set("spark.memory.offHeap.size","64g") 
sc = SparkContext(conf = config)

reg = L1L2(l1 = 0.01, l2 = 0.01)
model = ml.Conv_LSTM(num_classes = 2, timesteps = 8, reg = reg)
model.summary()
convLSTM = ml.model_train(model, number_epoch = 20)

KNN = KNeighborsClassifier(n_neighbors = 3)
NB = GaussianNB()
SVM = svm.SVC()
LR = linear_model.LogisticRegression(C = 0.01)
RF = ensemble.RandomForestClassifier(n_estimators = 5, n_jobs = -1)
GBT = ensemble.GradientBoostingClassifier(n_estimators = 5, max_leaf_nodes = 3, max_depth = 3)

clfs = [('LR',LR),('SVM',SVM),('NB',NB),('KNN',KNN),('GBT',GBT),('RF',LR),('Conv-LSTM Regression',convLSTM)]


def ddiDataGenerator(clfs, ddi_list, vectors):
    ddi_df = pd.read_csv(ddi_list, sep='\t')

    embedding_df = pd.read_csv(vectors, delimiter = '\t') 

    embedding_df.Entity = embedding_df.Entity.str[-8:-1]
    embedding_df.rename(columns = {'Entity':'Drug'}, inplace = True)

    len(set(ddi_df.Drug1.unique()).union(ddi_df.Drug2.unique()) )
    pairs, classes = ml.generatePairs(ddi_df, embedding_df)
    
    return pairs, classes, embedding_df

ddi_list = 'full_DDI.txt'
vectors = 'RDF2Vec_sg_300_5_5_15_2_500_d5_uniform.txt'

pairs, classes, embedding_df = ddiDataGenerator(clfs, ddi_list, vectors)

# Hyperparameters 
n_seed = 100
n_fold = 5 
n_run = 10
n_proportion = 1

scoreDF = ml.kfoldCV(sc, pairs, classes, embedding_df, clfs, n_run, n_fold, n_proportion, n_seed)

scoreDF.groupby(['method','run']).mean().groupby('method').mean()
scoreDF.to_csv('Results.txt',sep = '\t', index = False)