def test_one_batch_fit_transform():
n_samples = 1000
n_features = 100
data = np.random.randn(n_samples, n_features).astype(DTYPE)
# -- reference outputs
ref_mean = data.mean(axis=0)
ref_std = data.std(axis=0)
ref_out = (data - ref_mean[np.newaxis]) / ref_std[np.newaxis, :]
# -- calculation with OnlineScaler
scaler = OnlineScaler(n_features)
out = scaler.fit_transform(data)
# -- tests
assert_allclose(out, ref_out, rtol=RTOL, atol=ATOL)
def test_multiple_batch_fit_for_mean_and_std():
n_samples = 1000
n_features = 100
n_batches = 5
data = np.random.randn(n_batches * n_samples, n_features).astype(DTYPE)
# -- reference outputs
ref_mean = data.mean(axis=0)
ref_std = data.std(axis=0)
# -- calculation with OnlineScaler
scaler = OnlineScaler(n_features)
for i in xrange(n_batches):
scaler.partial_fit(data[(i * n_samples):((i + 1) * n_samples), :])
out_mean = scaler.r_mn
out_std = np.sqrt(scaler.r_va)
# -- tests
assert_allclose(out_mean, ref_mean, rtol=RTOL, atol=ATOL)
assert_allclose(out_std, ref_std, rtol=RTOL, atol=ATOL)