experimentsClassicClassifiers.py

import sys
import os
import time
import argparse
import pandas as pd
import numpy as np
from data_preprocessing import Preprocessor
from features_extractor import feature_extractor
from ClassPreprocessor import remove_weak_classes, resample_classes, create_dict, split_class
from classifiers import classifiers, grid_params
from sklearn.model_selection import train_test_split
from sklearn import preprocessing
from evaluate_model import evaluate_model
from sklearn.model_selection import GridSearchCV
import configparser
import pickle

import nltk
nltk.download('stopwords')
nltk.download('punkt')

parser = argparse.ArgumentParser()
parser.add_argument("dataPath", help="Path of the dataframe")
parser.add_argument("columnText", help="the column name of the text that should preproceed", default = 'content')
parser.add_argument("columnClass", help="ColumnClass the column name of the classes")
parser.add_argument("minOfInstancePerClass", help="minOfInstancePerClass the minimum of instance required for each class", type=int)
parser.add_argument("maxOfInstancePerClass", help="maxOfInstancePerClass the maximum of instance required resamling classes", type=int)

args = parser.parse_args()
dataPath = args.dataPath
columnText = args.columnText
columnClass = args.columnClass
minOfInstancePerClass = args.minOfInstancePerClass
maxOfInstancePerClass = args.maxOfInstancePerClass

if not os.path.exists('reports'):
    os.makedirs('reports')

if not os.path.exists(os.path.join('reports',  columnClass)):
    os.makedirs(os.path.join('reports', columnClass))

# create directory in the reports directory so save the classification results
dir_name_report = str(minOfInstancePerClass) + '_' + str(maxOfInstancePerClass)
if not os.path.exists(os.path.join('reports',  columnClass, dir_name_report)):
    os.makedirs(os.path.join('reports', columnClass, dir_name_report))


# create directory to save and load models
if not os.path.exists('models'):
    os.makedirs('models')

# Reading data and preprocessings steps
preprocessor = Preprocessor()

df = pd.read_csv(dataPath, sep="\t")

df = remove_weak_classes(df, columnClass, minOfInstancePerClass)
df = resample_classes(df, columnClass, maxOfInstancePerClass)

#Read configuration file for retreiving parameters of features extractors

config = configparser.ConfigParser()
config.read('settings.conf')

vectorization_max_df = int(config.get('vectorizers','vectorization_max_df')) if config.get('vectorizers','vectorization_max_df').isdigit() else  float(config.get('vectorizers','vectorization_max_df'))
vectorization_min_df = int(config.get('vectorizers','vectorization_min_df')) if config.get('vectorizers','vectorization_min_df').isdigit() else  float(config.get('vectorizers','vectorization_min_df'))
vectorization_numberOfFeatures = int(config.get('vectorizers','vectorization_numberOfFeatures')) if config.get('vectorizers','vectorization_numberOfFeatures').isdigit() else None

doc2vec_vec_size = int(config.get('vectorizers','doc2vec_vec_size'))
max_epochs = int(config.get('vectorizers','max_epochs'))
doc2vec_min_count = int(config.get('vectorizers','doc2vec_min_count'))
doc2vec_dm = int(config.get('vectorizers','doc2vec_dm')) # If dm=1, ‘distributed memory’ (PV-DM) is used. Otherwise, distributed bag of words (PV-DBOW) is employed.
doc2vec_workers = int(config.get('vectorizers','doc2vec_workers'))

print("size after resampling, ",len(df))

#prepare data
#df = df[df[columnClass] != 'unclassified']
y  = df[columnClass]

print(df.head())

print(df[columnClass].head())

train_x, test_x, train_y, test_y = train_test_split(df, y, test_size=0.33, random_state=42, stratify = y )
encoder = preprocessing.LabelEncoder()
train_y = encoder.fit_transform(train_y)
valid_y = encoder.fit_transform(test_y)

print("size training set, ",len(train_x))
print("size validation set, ",len(test_x))


for columnInput in [columnText, 'firstParagraph']:

    print('Process: ' + columnInput)

    extractor = feature_extractor(train_x, test_x, columnInput, columnClass)

    features_techniques = [
    ('counter',  extractor.count_vect(max_df = vectorization_max_df, min_df = vectorization_min_df, numberOfFeatures = vectorization_numberOfFeatures )),
    ('tf_idf',  extractor.tf_idf(max_df = vectorization_max_df, min_df = vectorization_min_df, numberOfFeatures = vectorization_numberOfFeatures)),
    ('doc2vec',  extractor.doc2vec(max_epochs, doc2vec_vec_size, doc2vec_min_count ,  doc2vec_dm))]



    #case of full text
    for feature_technique_name, features in features_techniques:

        print("**** Classifier :", feature_technique_name)
        # features has the train_x and the test_x after vectorization
        train_x, test_x = features

        for tmp_clf, tmp_grid_params in zip(classifiers, grid_params):
            clf_name, clf = tmp_clf
            grid_param_name, grid_param = tmp_grid_params
            print(clf_name, clf, grid_param_name, grid_param)
            model_file_name = columnInput + '_' +feature_technique_name + '_' + clf_name+ str(minOfInstancePerClass) + '_' + str(maxOfInstancePerClass) +".pkl"

            if clf_name != 'bayes' :
                clf = GridSearchCV(clf, grid_param, refit = True, verbose = 3, n_jobs=-1)
            elif feature_technique_name == 'doc2vec':
                    continue

            t_begin = time.time()

            if os.path.isfile(os.path.join('./models', model_file_name)):
                report, accuracy, weighted_avg = evaluate_model(clf, test_x, valid_y, y_pred, valid_y, [str(e) for e in encoder.transform(encoder.classes_)],  encoder.classes_, os.path.join('reports', columnClass, dir_name_report, file_name_report)+'.pdf')
                with open(os.path.join('./models',columnClass, model_file_name), 'rb') as file:
                    clf = pickle.load(file)
            else:
                with open(os.path.join('./models',columnClass, model_file_name), 'wb') as file:
                    clf.fit(train_x, train_y)
                    pickle.dump(clf, file)

            t_end =time.time()

            training_time = t_end - t_begin

            y_pred = clf.predict(test_x)

    #evaluate model
            file_name_report = columnInput + '_' +feature_technique_name + '_' + clf_name

            report, accuracy, weighted_avg = evaluate_model(clf, test_x, valid_y, y_pred, valid_y, [str(e) for e in encoder.transform(encoder.classes_)],  encoder.classes_, os.path.join('reports', columnClass, dir_name_report, file_name_report)+'.pdf')
            report.to_csv(os.path.join('reports', columnClass, dir_name_report, file_name_report+'.csv'))
            with open(os.path.join('reports', columnClass, dir_name_report, file_name_report+'.txt'), 'w') as f:

                sys.stdout = f # Change the standard output to the file we created.
                print('accuracy : {}'.format(accuracy))
                print('weighted_Precision : {}'.format(weighted_avg['precision']))
                print('weighted_Recall    : {}'.format(weighted_avg['recall']))
                print('weighted_F-score   : {}'.format(weighted_avg['f1-score']))
                print('weighted_Support   : {}'.format(weighted_avg['support']))
                print(dict(zip(encoder.classes_, encoder.transform(encoder.classes_))))
                print('training time   : {}'.format(training_time))
                try:
                    print('best parameters   : {}'.format(clf.best_params_))
                except AttributeError:
                    pass

                #sys.stdout = sys.stdout # Reset the standard output to its original value
                sys.stdout = sys.__stdout__