Add accuracy printout and use pipelined flow

Author: maillouxc

part2/experiments.py

@@ -3,28 +3,44 @@
 import argparse
 import os
 import glob
+import numpy
 from time import time
-
 from sklearn.feature_extraction.text import CountVectorizer
 from sklearn.feature_extraction.text import TfidfTransformer
+from sklearn.pipeline import Pipeline
 from sklearn import svm
 from sklearn.linear_model import Perceptron
 
 
 def main():
  parsed_args = parse_command_line_arguments()
+ 
  data_path = parsed_args.path
-
- # We need the tfidf_transformer and counts_vect here to pass the the test email loader.
- # There are other ways to do this, but this is the most convenient.
- # It's important to supply this tranformer because we need to maintain the vocabulary between test and training
- training_emails, training_labels, tfidf_transformer, counts_vect = load_training_dataset(data_path + "/training")
- test_emails, test_labels = load_test_dataset(data_path + "/test", tfidf_transformer, counts_vect)
-
- svm_classifier = train_svm_classifier(training_emails, training_labels)
-
+ training_emails, training_labels, = load_dataset(data_path + "/training")
+ test_emails, test_labels = load_dataset(data_path + "/test")
+
+ svm_classifier = Pipeline([
+ ('count_vectorizer', CountVectorizer(stop_words="english")),
+ ('tfidf_transformer', TfidfTransformer()),
+ ('svm_classifier', svm.SVC(gamma="scale",
+ C=1,
+ verbose=False))
+ ])
+
+ # Train the svm
+ svm_classifier.fit(training_emails, training_labels)
+
+ num_correctly_classified = 0
+ num_wrongly_classified = 0
+ # Test it
  results = svm_classifier.predict(test_emails)
- print(results)
+ for i in range(len(results)):
+ if (results[i] == test_labels[i]):
+ num_correctly_classified += 1
+ else:
+ num_wrongly_classified += 1
+
+ print("Accuracy of SVM classifier was " + str("{0:.3%}").format(numpy.mean(results == test_labels)))
 
 
 def parse_command_line_arguments():
@@ -37,7 +53,7 @@ def parse_command_line_arguments():
  return parser.parse_args()
 
 
-def load_training_dataset(data_path):
+def load_dataset(data_path):
  print("Loading training dataset at " + data_path)
 
  email_data = []
@@ -52,46 +68,9 @@ def load_training_dataset(data_path):
  with open(file, 'r') as email_file:
  email_data.append(email_file.read())
 
- count_vectorizer = CountVectorizer(stop_words='english')
- tfidf_transformer = TfidfTransformer(use_idf=False)
- email_feature_data = tfidf_transformer.fit_transform(count_vectorizer.fit_transform(email_data))
-
- print(email_feature_data)
  print("Loaded " + str(len(email_labels)) + " training emails")
 
- return email_feature_data, email_labels, tfidf_transformer, count_vectorizer
-
-
-def load_test_dataset(data_path, tfidf_transformer, counts_vect):
- print("Loading test dataset at " + data_path)
-
- email_data = []
- email_labels = []
-
- os.chdir(data_path)
- for file in os.listdir():
- if (file.startswith("sp")):
- email_labels.append("spam")
- else:
- email_labels.append("ham")
- with open(file, 'r') as email_file:
- email_data.append(email_file.read())
-
- email_feature_data = tfidf_transformer.transform(counts_vect.transform(email_data))
-
- print(email_feature_data)
- print("Loaded " + str(len(email_labels)) + " test emails")
-
- return email_feature_data, email_labels
-
-
-def train_svm_classifier(emails, labels):
- classifier = svm.SVC(
- gamma='scale',
- C=1,
- verbose=True)
- classifier.fit(emails, labels)
- return classifier
+ return email_data, email_labels
 
 
 if __name__ == "__main__":