thesis-anne/src/SVM_multiclass.py

'''
Support Vector Machines (SVM) Classifier
========================================

The SVM training algorithm builds a model from the training data that assigns
the test samples to one category ('merger' or 'not merger'),
making it a non-probabilistic binary linear classifier.
An SVM model is a representation of the samples as points in space,
mapped so that the examples of the separate categories are divided
by a clear gap that is as wide as possible.
New samples are then mapped into that same space and predicted
to belong to a category based on which side of the gap they fall.
'''

from BagOfWords import BagOfWords

import csv

import pandas as pd
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.feature_selection import SelectPercentile
from sklearn.metrics import recall_score, precision_score, f1_score, make_scorer, accuracy_score
from sklearn.model_selection import StratifiedKFold
from sklearn.model_selection import GridSearchCV
from sklearn.pipeline import Pipeline
from sklearn.svm import LinearSVC
from sklearn.svm import SVC
from sklearn.svm import NuSVC

class SVM_multiclass:

	def make_svm(dataset, sklearn_cv=True, percentile=100):

		print('# starting multinomial svm')
		print('# ...')

		# split data into text and label set
		# join title and text
		X = dataset['Title'] + '. ' + dataset['Text']
		y = dataset['Label']

		if sklearn_cv:

			# ignore company names
			company_names_list = BagOfWords.load_company_names()
			stopwords = list(BagOfWords.set_stop_words()).extend(company_names_list)
			cv = CountVectorizer(stop_words = stopwords)

		# use stratified k-fold cross-validation as split method
		skf = StratifiedKFold(n_splits = 10, shuffle=True, random_state=5)

		classifier = LinearSVC()
		
		# for predict proba:
		#classifier = SVC(probability=True,
		#					 gamma='auto')

		# metrics
		recall_scores = []
		precision_scores = []
		accuracy_scores = []
		f1_scores = []

		# for each fold
		n = 0
		for train, test in skf.split(X,y):

			n += 1
			print('# split no. ' + str(n))

			if sklearn_cv:
				# use sklearn CountVectorizer
				# fit the training data and then return the matrix
				
				training_data = cv.fit_transform(X[train], y[train]).toarray()
				# transform testing data and return the matrix
				testing_data = cv.transform(X[test]).toarray()
			else:
				# use my own BagOfWords python implementation
				stemming = True
				rel_freq = True
				extracted_words = BagOfWords.extract_all_words(X[train])
				vocab = BagOfWords.make_vocab(extracted_words)

				# fit the training data and then return the matrix
				training_data = BagOfWords.make_matrix(extracted_words,
								vocab, rel_freq, stemming)
				# transform testing data and return the matrix
				extracted_words = BagOfWords.extract_all_words(X[test])
				testing_data = BagOfWords.make_matrix(extracted_words,
								vocab, rel_freq, stemming)

			# apply select percentile
			selector = SelectPercentile(percentile=percentile)
			selector.fit(training_data, y[train])

			# new reduced data sets
			training_data_r = selector.transform(training_data)
			testing_data_r = selector.transform(testing_data)

			#fit classifier
			classifier.fit(training_data_r, y[train])
			#predict class
			predictions_train = classifier.predict(training_data_r)
			predictions_test = classifier.predict(testing_data_r)

			#print and store metrics
			rec = recall_score(y[test], predictions_test, average='weighted')
			print('rec: ' + str(rec))
			recall_scores.append(rec)
			prec = precision_score(y[test], predictions_test, average='weighted')
			print('prec: ' + str(prec))
			print('#')
			precision_scores.append(prec)
			acc = recall_score(y[test], predictions_test, average='weighted')
			accuracy_scores.append(acc)
			print('acc: ' + str(acc))
			print('#')
			# equation for f1 score
			f1_scores.append(2 * (prec * rec)/(prec + rec))

			#class_prob.append(classifier.class_prior_)
			#class_counts.append(classifier.class_count_)
			#print(classifier.predict_proba(testing_data_r))

		##########################
		# classes in order used
		classes = classifier.classes_
		
		print('Recall (Min): ' + str(min(recall_scores)))
		print('Recall (Max): ' + str(max(recall_scores)))
		print('Recall (Average): ' + str(sum(recall_scores)/len(recall_scores)))
		print()
		print('Precision (Min): ' + str(min(precision_scores)))
		print('Precision (Max): ' + str(max(precision_scores)))
		print('Precision (Average) :' + str(sum(precision_scores)/len(precision_scores)))
		print()
		print('Accuracy (Min): ' + str(min(accuracy_scores)))
		print('Accuracy (Max): ' + str(max(accuracy_scores)))
		print('Accuracy (Average) :' + str(sum(accuracy_scores)/len(accuracy_scores)))

		# return classes and vector of class estimates
		return recall_scores, precision_scores

if __name__ == '__main__':

	# read csv file
	print('# reading dataset')
	print('# ...')

	# read current data set from csv
	df = pd.read_csv('../data/interactive_labeling_round_11.csv',
			  sep='|',
			  usecols=range(1,13), # drop first column 'unnamed'
			  encoding='utf-8',
			  quoting=csv.QUOTE_NONNUMERIC,
			  quotechar='\'')

	# select only labeled articles
	SVM_multiclass.make_svm(df.loc[df['Label'] != -1].reset_index(drop=True), 
							sklearn_cv=True)