def _prepare_X(self, X_raw):
X_raw, self.bad_movie_indexes = remove_bad_movies(
X_raw, self.bad_movie_indexes)
X, _ = fill_ratings_with_mean_per_user(X_raw)
X = X[:, 1, :] # only ratings
return X.reshape(X.shape[0], -1)
def _prepare_X(self, X_raw, train = True):
X_raw, self.bad_movie_indexes = remove_bad_movies(X_raw, self.bad_movie_indexes)
X, _ = fill_ratings_with_mean_per_user(X_raw)
X = X[:,1,:] # only ratings
X = X.reshape(X.shape[0], -1)
if train:
self.release_date = np.min(X_raw[:,2,0])
X = np.hstack((X, diff_in_time_to_rate_target_movie(X_raw, self.release_date)))
return X
def _prepare_X(self, X_from_input):
X, _ = fill_ratings_with_mean_per_user(X_from_input)
X = X[:, 1, :] # only ratings
if self.bad_movie_indexes is None:
n_users_didnt_see_movie = np.sum(np.isnan(X_from_input[:, 1, :]),
axis=0)
self.bad_movie_indexes = np.argwhere(
n_users_didnt_see_movie > 2500).squeeze()
X = np.delete(X, self.bad_movie_indexes, axis=1)
return X.reshape(X.shape[0], -1)
def _prepare_X(self, X_raw, Y=None, train=False):
X_raw, self.bad_movie_indexes = remove_bad_movies(
X_raw, self.bad_movie_indexes)
X, _ = fill_ratings_with_mean_per_user(X_raw)
X = X[:, 1, :] # only ratings
if train:
self.scaler.fit(X)
X_rescaled = self.scaler.transform(X)
if train:
self.pca = PCA(n_components=75).fit(X_rescaled)
X_components = self.pca.transform(X_rescaled)
return X_components.reshape(X.shape[0], -1)
def _prepare_X(self, X_from_input):
X, _ = fill_ratings_with_mean_per_user(X_from_input)
X = X[:,1,:] # only ratings
return X.reshape(X.shape[0], -1)
