def _post_train_calls(self):
""" Function calls to perform after training, such as computing eigenvectors """
if self.components > 0:
self.svd_res = compute_principal_components(
self.sv.vectors, components=self.components)
remove_principal_components(self.sv.vectors,
svd_res=self.svd_res,
inplace=True)
else:
self.svd_res = 0
logger.info(f"no removal of principal components")
def _post_train_calls(self):
""" Function calls to perform after training, such as computing eigenvectors """
if self.components > 0:
self.svd_res = compute_principal_components(
self.sv.vectors,
components=self.components,
cache_size_gb=self.cache_size_gb,
)
self.svd_weights = (self.svd_res[0]**2) / (self.svd_res[0]**
2).sum().astype(REAL)
remove_principal_components(
self.sv.vectors,
svd_res=self.svd_res,
weights=self.svd_weights,
inplace=True,
)
else:
self.svd_res = 0
logger.info(f"no removal of principal components")
def test_remove_weighted_components(self):
m = np.ones((500,10), dtype=np.float32)
out = compute_principal_components(vectors = m)
remove_principal_components(m, svd_res=out, weights=np.array([0.5]))
self.assertTrue(np.allclose(0.75, m))
def test_remove_components(self):
m = np.ones((500,10), dtype=np.float32)
out = compute_principal_components(vectors = m)
res = remove_principal_components(m, svd_res=out, inplace=False)
self.assertTrue(np.allclose(1., res, atol=1e-5))
def test_compute_large_components(self):
m = np.random.uniform(size=(int(2e6), 100)).astype(np.float32)
out = compute_principal_components(vectors=m, cache_size_gb=0.2)
self.assertEqual(2, len(out))
self.assertEqual(1, len(out[1]))
self.assertEqual(np.float32, out[1].dtype)