def to_libfm(examples, libfm_filename):
libfm_file = open(libfm_filename, 'w')
for example in examples:
x_i = model.represent(example)[:-1]
y_i = model.label(example)
user_id = example['user']
user_id_feature = "%d:1" % (len(x_i)+user_id)
libfm_file.write("%d %s %s\n" % (y_i, sp_vector(x_i), user_id_feature))
libfm_file.close()
def to_libfm(examples, libfm_filename):
libfm_file = open(libfm_filename, 'w')
for example in examples:
x_i = model.represent(example)[:-1]
y_i = model.label(example)
user_id = example['user']
user_id_feature = "%d:1" % (len(x_i) + user_id)
libfm_file.write("%d %s %s\n" % (y_i, sp_vector(x_i), user_id_feature))
libfm_file.close()
def get_test_examples():
global test_X
global test_y
if not test_X:
print "Loading test examples"
_log_time()
test_examples = music.load_examples("data/test_40k_10k.pkl")
_print_time_diff()
print "Obtaining X and y values"
_log_time()
test_X = [model.represent(example) for example in test_examples]
_print_time_diff()
_log_time()
test_y = [model.label(example) for example in test_examples]
_print_time_diff()
return test_X, test_y
print "Readying testing data"
_log_time()
music.ready_testing_data()
_print_time_diff()
_draw_separator()
print "Start loading examples"
_log_time()
examples = music.load_examples("data/train.pkl")
_print_time_diff()
_draw_separator()
print "Obtaining all x and y values"
_log_time()
all_X = [model.represent(example) for example in examples]
_print_time_diff()
_log_time()
all_y = [model.label(example) for example in examples]
_print_time_diff()
_draw_separator()
def print_consolidated_scores(scores):
_draw_separator(".", 5)
print "%0.6f (+/- %0.6f)" % (scores.mean(), scores.std() / 2)
_draw_separator("`", 5)
def _get_mean_score(scores):
return "%0.6f" % (scores.mean())
scores = cross_val_score(knr, X, y, scoring='neg_mean_squared_error', cv=3)
return scores
if __name__ == "__main__":
import music
train_examples = music.load_examples('data/train.pkl')
# poly = PolynomialNetworkRegressor(degreex=3, n_components=2, tol=1e-3, warm_start=True, random_state=0)
fm = pylibfm.FM(num_iter=10,
verbose=True,
task="regression",
initial_learning_rate=0.001,
learning_rate_schedule="optimal")
v = DictVectorizer()
X = np.asarray([model.represent(example) for example in train_examples])
y = np.asarray([model.label(example) for example in train_examples])
# fm.fit(sparse.csr_matrix(X), y)
# svr_rbf.fit(X, y)
# pca = PCA(n_components=100)
# pca.fit(X)
# X_fit = pca.transform(X)
# print "pca done"
# xs = [x[0] for x in X_fit]
# ys = [x[1] for x in X_fit]
# plt.scatter(xs, ys)
# plt.show()
# print v.fit_transform(X)
# print X_fitM
y_np = np.asarray(y)
plt.hist(y_np, bins=np.arange(y_np.min(), y_np.max() + 1))