def knn_experiment():
K = len(topics)
dm = data.create_data_manager()
ordered_topics = dm.order_topics(topics)
print 'loading train data . . .'
X_train, Y_train = dm.load_data('train')
Y_train_slice = dm.slice_Y(Y_train, ordered_topics)
print 'loading test data . . .'
X_test, Y_test = dm.load_data('test')
Y_gold = dm.slice_Y(Y_test, ordered_topics)
final_results = {}
print 'interating over models . . .'
for k in [1,15,31,45,61]:
model = '%d-nearest neighbours' % k
print 'now using model %s . . .' % model
learner = knn.MultikNN(K, k)
# K being the number of topics, and k being the number of neighbours
learner.train(X_train, Y_train_slice)
Y_pred = learner.batch_predict_classes(X_test)
results = evaluate.per_topic_results(Y_pred, Y_gold)
results_dict = {topic: result for (topic, result) in zip(ordered_topics, results)}
pprint(results_dict)
final_results[model] = results_dict
print 'saving final results . . .'
with open('results/knn_final_results.txt', 'w') as f:
pprint(final_results, stream=f)
def distance_matrix(self):
print 'creating distance_matrix . . .'
dm = data.create_data_manager()
X_train, _ = dm.load_data('train')
X_test, _ = dm.load_data('test')
distances = np.empty((X_test.shape[0], X_train.shape[0]))
for row in xrange(X_test.shape[0]):
print 'row', row
x = X_test[row]
distances[row] = np.array([self._distance(x, X_train[col]) for col in xrange(X_train.shape[0])])
np.save('distance_matrix.npy', distances)
return distances
def test2():
print 'loading data . . .'
dm = data.create_data_manager()
X_train, Y_train = dm.load_data('train')
Y_train_slice = dm.slice_Y(Y_train, topics)
print 'training knn . . .'
K = len(topics)
learner = knn.MultikNN(K, 30)
learner.train(X_train, Y_train_slice)
print 'testing knn . . .'
X_test, Y_test = dm.load_data('test')
Y_test_slice = dm.slice_Y(Y_test, topics)
errors = learner.classification_errors(X_test, Y_test_slice)
return errors
def test():
# test run using degree-3 polynomial kernel with C=1
# for the ten main topics in reuters
dm = data.create_data_manager()
print 'loading train data . . .'
X_train, Y_train = dm.load_data('train')
Y_train_slice = dm.slice_Y(Y_train, topics)
print 'training svm . . .'
K = len(topics)
learner = svm2.MultiSVM(K, 1.0, 'poly', 3)
learner.train(X_train, Y_train_slice)
print 'loading test data . . .'
X_test, Y_test = dm.load_data('test')
Y_gold = dm.slice_Y(Y_test, topics)
print 'predicting train . . .'
Y_pred = learner.batch_predict_classes(X_train)
print 'evaluating . . .'
precision, recall = evaluate.precision_recall(Y_pred, Y_train_slice)
f1 = evaluate.f_score(precision, recall)
print 'Precision: %.3f' % precision
print 'Recall: %.3f' % recall
print 'F1: %.3f' % f1
print 'predicting test . . .'
Y_pred = learner.batch_predict_classes(X_test)
print 'evaluating . . .'
precision, recall = evaluate.precision_recall(Y_pred, Y_gold)
f1 = evaluate.f_score(precision, recall)
print 'Precision: %.3f' % precision
print 'Recall: %.3f' % recall
print 'F1: %.3f' % f1
def svm_experiment(C, balance=''):
K = len(topics)
dm = data.create_data_manager()
ordered_topics = dm.order_topics(topics)
print 'loading train data . . .'
X_train, Y_train = dm.load_data('train')
Y_train_slice = dm.slice_Y(Y_train, ordered_topics)
print 'loading test data . . .'
X_test, Y_test = dm.load_data('test')
Y_gold = dm.slice_Y(Y_test, ordered_topics)
final_results = {}
#print (Y_train_slice.sum(axis=0) + Y_train_slice.shape[0]).tolist()
#raise Exception
print 'interating over models . . .'
for model, kernel in svm_kernels:
print 'now using model %s . . .' % model
learner = svm.MultiSVM(K, C, kernel)
learner.train(X_train, Y_train_slice,
ordered_topics,
balance=balance, # use balance = '' if don't want to balance
max_per_class=3000)
Y_pred = learner.batch_predict_classes(X_test)
results = evaluate.per_topic_results(Y_pred, Y_gold)
results_dict = {topic: result for (topic, result) in zip(ordered_topics, results)}
pprint(results_dict)
final_results[model] = results_dict
print 'saving final results . . .'
with open('results/svm_final_results_C_%.1f_%s.txt' % (C,balance), 'w') as f:
pprint(final_results, stream=f)
def __init__(self, XY=None):
self.dm = data.create_data_manager()
if XY == None:
self.X, self.Y = self.dm.load_data('train')
else:
self.X, self.Y = XY