print np.var([i[0] for i in negative]) print np.var([i[1] for i in negative]) print np.var([i[0] for i in positive]) print np.var([i[1] for i in positive]) print "----" clustering = KMeans(n_clusters=4, n_jobs=-2, n_init=12, init='random') clustering.fit(training_data) test_data = pca.transform(test_data) a = clustering.predict(test_data) scores = gf.precision_recall_etc(a, test_category) print scores """ a_positive = a[:150] a_negative = a[150:] print len([i for i in a_positive if i==0]) print len([i for i in a_positive if i==1]) print len([i for i in a_positive if i==2]) print len([i for i in a_positive if i==3]) #print len([i for i in a_positive if i==4]) print "-" print len([i for i in a_negative if i==0]) print len([i for i in a_negative if i==1]) print len([i for i in a_negative if i==2]) print len([i for i in a_negative if i==3])
# Loop over different step sizes to see if "fine tuning" the step size matters
for val in steps_to_try:
# Define a feature selector
feature_sel = RFECV(estimator_to_use, step=val, cv=8, n_jobs=3, verbose=1)
# Define a pipeline to feature select and classify the data, then train
pipeline = Pipeline([('select', feature_sel), ('berno', classifier)])
pipeline.fit(bagofwords_training, sentiment_training)
# Predict sentiment of test data
predict_test = pipeline.predict(bagofwords_test)
# Calculate performance and number of features, then collect and store these values
num_features.append(pipeline.named_steps['select'].n_features_)
output = general_f.precision_recall_etc(predict_test, sentiment_test)
precision.append(output['precision'])
recall.append(output['recall'])
specificity.append(output['specificity'])
NPV.append(output['NPV'])
f1.append(output['f1'])
# Plot out performance as a function of number of features
print "precision max: " + str(np.argmax(precision))
plt.plot(num_features, precision, label='precision')
print "recall max: " + str(np.argmax(recall))
plt.plot(num_features, recall, label='recall')
print "spec. max: " + str(np.argmax(specificity))
plt.plot(num_features, specificity, label='spec.')
print "NPV max: " + str(np.argmax(NPV))
plt.plot(num_features, NPV, label='NPV')
sentiment_training = np.loadtxt('output/out' + ignore + '_classes_' + number +
'.txt',
unpack=True)
# Read in the bag of words representation of the data
bagofwords_training = general_f.read_bagofwords_dat('output/out' + ignore +
'_bag_of_words_' + number +
'.csv')
# The same for the test data
sentiment_test = np.loadtxt('output/test' + ignore + '_classes_' + number +
'.txt',
unpack=True)
bagofwords_test = general_f.read_bagofwords_dat('output/test' + ignore +
'_bag_of_words_' + number +
'.csv')
# Define and train the classifier
classifier = RFC(n_estimators=150, criterion='entropy', n_jobs=3,
verbose=0).fit(bagofwords_training, sentiment_training)
# Predict sentiment of the test data
predict_test = classifier.predict(bagofwords_test)
# Calculate accuracy and print
print ""
test_percentage = general_f.accuracy_percentage(predict_test, sentiment_test)
print "RFC Test percentage for " + number + " : " + str(test_percentage)
print 'other stuff of test data: '
print general_f.precision_recall_etc(predict_test, sentiment_test)