def randomly_project_gradients_and_hessians(self,
gradients,
hessians,
y,
random_state=None):
# gradients = fillnan(gradients)
try:
proj_g = SparseRandomProjection(
n_components=1,
random_state=random_state,
dense_output=True,
).fit_transform(X=gradients, y=y)
except ValueError:
warnings.warn('Couldn\'t project the gradients so applied fillnan')
gradients = np.nan_to_num(gradients)
proj_g = SparseRandomProjection(
n_components=1,
random_state=random.randint(1, 1000),
dense_output=True,
).fit_transform(X=gradients, y=y)
proj_h = hessians #SparseRandomProjection(n_components=1, random_state=self.random_state).fit_transform(X=hessians)
return proj_g.ravel().astype(np.float32), proj_h.astype(np.float32)
def randomly_project_gradients_and_hessians(self, gradients, hessians):
proj_g = SparseRandomProjection(
n_components=1,
random_state=self.random_state).fit_transform(X=gradients)
proj_h = hessians #SparseRandomProjection(n_components=1, random_state=self.random_state).fit_transform(X=hessians)
return proj_g.ravel().astype(np.float32), proj_h.astype(np.float32)
