def test_fit_predict_normalize():
'''Tests with dataset'''
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2)
ssm = SSM(False, False, 1, normalize_err=True)
Y = ssm.fit_predict(D)
assert Y.any()
def test_yes():
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2).np_like_firstn()
tw = TemporalWeight('yes', 2)
y = tw.fit_predict(D)
for i in xrange(D.shape[0]):
val = D[i][-1]
assert_equal(val, y[i])
def test_yes():
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2).np_like_firstn()
tw = TemporalWeight('yes', 2)
y = tw.fit_predict(D)
for i in xrange(D.shape[0]):
val = D[i][-1]
assert_equal(val, y[i])
def test_rbf_with_dataset():
'''Tests the RBF model with TimeSeriesDataset'''
rbf_model = RidgeRBFModel(10, .5, alpha=.01)
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-6)
y = D.np_like_firstn().sum(axis=1)
model = rbf_model.fit(D, y)
y_pred = model.predict(D)
mrse = (((y - y_pred) / y)**2).mean()
assert_almost_equal(0, mrse, 4)
def test_rbf_with_dataset():
'''Tests the RBF model with TimeSeriesDataset'''
rbf_model = RidgeRBFModel(10, .5, alpha=.01)
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-6)
y = D.np_like_firstn().sum(axis=1)
model = rbf_model.fit(D, y)
y_pred = model.predict(D)
mrse = (((y - y_pred) / y)**2).mean()
assert_almost_equal(0, mrse, 4)
def test_pow():
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2).np_like_firstn()
tw = TemporalWeight('pow', 2)
y = tw.fit_predict(D)
for i in xrange(D.shape[0]):
div = sum((np.arange(D.shape[1]) + 1)**2)
val = sum(D[i] * ((np.arange(D.shape[1]) + 1)**2) / div)
assert_equal(val, y[i])
def test_pow():
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2).np_like_firstn()
tw = TemporalWeight('pow', 2)
y = tw.fit_predict(D)
for i in xrange(D.shape[0]):
div = sum((np.arange(D.shape[1]) + 1) ** 2)
val = sum(D[i] * ((np.arange(D.shape[1]) + 1) ** 2) / div)
assert_equal(val, y[i])
def test_one_series():
'''Tests with only one time series'''
y = tsio.from_id_row_mat(SM_ONE_SEQUENCE, add_eps=1e-2).np_like_firstn()
pred = SpikeM(0).fit_predict(y, [24], True)
from matplotlib import pyplot as plt
x = range(y.shape[1])
plt.plot(x, y[0])
plt.plot(x, pred[0])
plt.show()
rmse = np.sqrt((pred[0] - y[0]) ** 2).sum() / len(y[0])
print(rmse)
assert rmse < 8
def test_one_series():
'''Tests with only one time series'''
y = tsio.from_id_row_mat(SM_ONE_SEQUENCE, add_eps=1e-2).np_like_firstn()
pred = SpikeM(0).fit_predict(y, [24], True)
from matplotlib import pyplot as plt
x = range(y.shape[1])
plt.plot(x, y[0])
plt.plot(x, pred[0])
plt.show()
rmse = np.sqrt((pred[0] - y[0])**2).sum() / len(y[0])
print(rmse)
assert rmse < 8
def test_fit_predict2():
'''
Tests the fit and predict methods for the state space models. Period Only
'''
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2).np_like_firstn()
D = D[: 50]
ssm = SSM(False, True, 50)
X_train = D[:, :50]
X_test = D[:, 50:]
Y = ssm.fit_predict(X_train)
err = ((X_test - Y) ** 2).mean()
assert_equal((50, 50), Y.shape)
assert err > 0
def test_fit_predict():
'''
Tests the fit and predict methods for the spike m
'''
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2).np_like_firstn()
D = D[:50]
sm = SpikeM()
X_train = D[:, :50]
X_test = D[:, 50:]
Y = sm.fit_predict(X_train)
err = ((X_test - Y)**2).mean()
assert_equal((50, ), Y.shape)
assert err > 0
def test_fit_predict():
'''
Tests the fit and predict methods for the spike m
'''
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2).np_like_firstn()
D = D[: 50]
sm = SpikeM()
X_train = D[:, :50]
X_test = D[:, 50:]
Y = sm.fit_predict(X_train)
err = ((X_test - Y) ** 2).mean()
assert_equal((50,), Y.shape)
assert err > 0
def test_ml_model():
'''Tests the ML model'''
ml_model = MLModel()
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-6).np_like_firstn()
D += 1e-6
X_train = D[:500, :7]
X_test = D[500:, :7]
y_train = D.sum(axis=1)[:500]
y_test = D.sum(axis=1)[500:]
model = ml_model.fit(X_train, y_train)
y_pred = model.predict(X_test)
mrse = (((y_test - y_pred) / y_test)**2).mean()
assert_equal(1, mrse > 0)
assert_equal(1, mrse < 1)
def test_rbf_model():
'''Tests the RBF model'''
rbf_model = RidgeRBFModel(100, 50, alpha=.00001)
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-6).np_like_firstn()
X_train = D[:500, :7]
X_test = D[500:, :7]
y_train = D.sum(axis=1)[:500]
y_test = D.sum(axis=1)[500:]
model = rbf_model.fit(X_train, y_train)
y_pred = model.predict(X_test)
mrse = (((y_test - y_pred) / y_test)**2).mean()
print(mrse)
assert_equal(1, mrse > 0)
assert_equal(1, mrse < 1)
def test_ml_model():
'''Tests the ML model'''
ml_model = MLModel()
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-6).np_like_firstn()
D += 1e-6
X_train = D[:500, :7]
X_test = D[500:, :7]
y_train = D.sum(axis=1)[:500]
y_test = D.sum(axis=1)[500:]
model = ml_model.fit(X_train, y_train)
y_pred = model.predict(X_test)
mrse = (((y_test - y_pred) / y_test)**2).mean()
assert_equal(1, mrse > 0)
assert_equal(1, mrse < 1)
def test_fit_predict():
'''
Tests the fit and predict methods for the state space models. Trend + period
'''
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2).np_like_firstn()
D = D[: 50]
ssm = SSM(True, True, 50)
D = (D.T / D.sum(axis=1)).T
X_train = D[:, :50]
X_test = D[:, 50:]
Y = ssm.fit_predict(X_train)
err = ((X_test - Y) ** 2).mean()
assert_equal((50, 50), Y.shape)
print(err)
assert err > 0
def test_rbf_model():
'''Tests the RBF model'''
rbf_model = RidgeRBFModel(100, 50, alpha=.00001)
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-6).np_like_firstn()
X_train = D[:500, :7]
X_test = D[500:, :7]
y_train = D.sum(axis=1)[:500]
y_test = D.sum(axis=1)[500:]
model = rbf_model.fit(X_train, y_train)
y_pred = model.predict(X_test)
mrse = (((y_test - y_pred) / y_test)**2).mean()
print(mrse)
assert_equal(1, mrse > 0)
assert_equal(1, mrse < 1)
def test_from_mat():
dataset = tsio.from_id_row_mat(YOUTUBE_1K)
assert_equal(1000, dataset.num_series)
for series in dataset:
assert_equal(100, len(series))
def test_from_mat():
dataset = tsio.from_id_row_mat(YOUTUBE_1K)
assert_equal(1000, dataset.num_series)
for series in dataset:
assert_equal(100, len(series))
def test_avg():
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2)
tw = TemporalWeight('avg')
y = tw.fit_predict(D)
assert_array_equal(D.np_like_firstn().mean(axis=1), y)
def test_avg():
D = tsio.from_id_row_mat(YOUTUBE_1K, add_eps=1e-2)
tw = TemporalWeight('avg')
y = tw.fit_predict(D)
assert_array_equal(D.np_like_firstn().mean(axis=1), y)
