def train_dev_test(w,
cv_data,
train_data,
train_labels,
dev_data,
test_data,
test_labels,
n_features,
name,
params,
dynamic=False,
average=False,
aggressive=False,
preprocessor=lambda x: x):
cv_acc, max_param = cross_validate(w, params, cv_data, dynamic, average,
aggressive)
max_w, train_acc = train_dev(w, max_param, train_data, train_labels,
dev_data[:, :-1], dev_data[:, -1], dynamic,
average, aggressive)
test_acc = test(max_w, test_data, test_labels)
def predictor(row):
label = np.sign(np.dot(row, max_w))
if label == -1:
label = 0
return label
write_output(name, max_param, cv_acc, train_acc, test_acc)
write_predictions(name[:3],
predictor,
n_features=n_features,
neg_labels=True,
bias=True,
preprocessor=preprocessor)
cv_acc = 0
if max_param is None:
cv_acc, max_param = cross_validate(cv_data, weights, update_weights,
params, update_params)
max_weights = train(train_data,
train_labels,
weights,
update_weights,
max_param,
update_params,
epochs=10)
train_acc = classify(train_data, train_labels, max_weights)
test_acc = classify(test_data, test_labels, max_weights)
def predictor(row):
label = np.sign(np.dot(row, max_weights))
if label == -1:
label = 0
return label
write_output('Logistic regression', max_param, cv_acc, train_acc, test_acc)
write_predictions('logreg',
predictor,
n_features=n_features,
neg_labels=True,
bias=True,
preprocessor=preprocessor)
cv_acc = 0
if max_param is None:
cv_acc, max_param = cross_validate(cv_data, weights, update_weights,
params, update_params)
max_weights = train(train_data,
train_labels,
weights,
update_weights,
max_param,
update_params,
epochs=100)
train_acc = classify(train_data, train_labels, max_weights)
test_acc = classify(test_data, test_labels, max_weights)
def predictor(row):
label = np.sign(np.dot(row, max_weights))
if label == -1:
label = 0
return label
write_output('SVM', max_param, cv_acc, train_acc, test_acc)
write_predictions('svm',
predictor,
n_features,
neg_labels=True,
bias=True,
preprocessor=preprocessor)
'../data/data-splits/data.test', n_features=n_features, preprocessor=preprocessor)
cv_data = np.array_split(np.hstack((train_data, train_labels)), 5)
max_acc = 0
opt_depth = 0
for i in range(2, n_features + 2):
acc = []
for j in range(len(cv_data)):
cv_test = cv_data[j]
cv_train = np.vstack(cv_data[:j] + cv_data[j + 1:])
tree, depth = id3(cv_train[:, :-1], cv_train[:, -1], max_depth=i)
cv_acc = evaluate_tree(cv_test[:, :-1], cv_test[:, -1], tree)
acc.append(cv_acc)
avg_acc = np.mean(acc)
if avg_acc > max_acc:
opt_depth = i
max_acc = avg_acc
tree, depth = id3(train_data, train_labels, max_depth=opt_depth)
train_acc = evaluate_tree(train_data, train_labels, tree)
test_acc = evaluate_tree(test_data, test_labels, tree)
write_output('ID3', opt_depth, max_acc, train_acc, test_acc)
write_predictions('id3', lambda row: classify(row, tree),
n_features=n_features, preprocessor=preprocessor)
for i in range(num_trees):
ri = np.random.choice(data_indices, size=data_size, replace=True)
rf = np.random.choice(feature_indices,
size=n_features - feature_size,
replace=False)
tree, depth = id3(train_data[ri],
train_labels[ri],
used_features=rf,
max_depth=max_depth,
split_size=split_size)
trees.append(tree)
if (i + 1) % (num_trees / 10) == 0:
print(str((i + 1) / num_trees * 100) + '%')
np.save('../data/trees', trees)
train_acc = evaluate_forest(train_data, train_labels, trees,
'../data/new_train')
test_acc = evaluate_forest(test_data, test_labels, trees,
'../data/new_test')
def predictor(row):
return mode(list(map(lambda tree: classify(row, tree), trees)))[0][0]
write_output('Random forest', None, None, train_acc, test_acc)
write_predictions('rf',
predictor,
n_features=n_features,
preprocessor=preprocessor)
lams = [2, 1.5, 1, 0.5]
train_data, train_labels = load_data(
'../data/data-splits/data.train', n_features=n_features, pos_labels=True)
test_data, test_labels = load_data(
'../data/data-splits/data.train', n_features=n_features, pos_labels=True)
cv_data = np.array_split(np.hstack((train_data, train_labels)), 5)
cv_acc, lam = cross_validate(cv_data, lams)
counts = count_pos(train_data, train_labels)
like, like_inv = naive_bayes(train_data.shape[0], counts, lam)
train_acc = classify(train_data, train_labels, like, like_inv)
test_acc = classify(test_data, test_labels, like, like_inv)
write_output('Naive Bayes', lam, cv_acc, train_acc, test_acc)
# Write predictions
with open('../data/data-splits/data.eval.id') as file:
eval_id = [int(line) for line in file]
eval_data, _ = load_data('../data/data-splits/data.eval.anon', n_features=n_features, pos_labels=True)
n_samples, n_features = eval_data.shape
samples_inv = np.ones((n_samples, n_features)) - eval_data
samples_inv = np.hstack((np.zeros((n_samples, 1)), samples_inv))
samples = np.hstack((np.ones((n_samples, 1)), eval_data))
probs = np.dot(samples, like) + np.dot(samples_inv, like_inv)
probs = probs.argmax(axis=1).reshape((-1, 1))