Steganografiya quyidagi sohalarda qo'llaniladi, lekin ular bilan cheklanmaydi

mode (str, optional): Classification or
regression mode

Returns:
ownership (dict): Watermark triggers information
X_train (array): Modified input data
y_train (array): Modified label data

"""
# Check if inputs are arrays
if type(X_train) != np.ndarray and type(y_train) != np.ndarray:
raise TypeError("Error: X and Y must be np.ndarray!")

WM_X, WM_y = np.array([]), np.array([])
ownership = {}
_, shape_x = X_train.shape

for k in range(self.trigger_size):
# Generate triggers
trigger_x = np.array([[random.random() for i in range(shape_x)]])
if mode == 'CLASSIFICATION':
trigger_y = np.array([random.choice(np.unique(y_train))])
else:
a, b = min(y_train), max(y_train)
trigger_y = np.array([random.randint(a, b)])
# Add instance to trigger set
if k == 0:
WM_X, WM_y = trigger_x, trigger_y
else:
WM_X = np.vstack((WM_X, trigger_x))
WM_y = np.vstack((WM_y, trigger_y))

# Add trigger to training data
X_train = np.vstack((X_train, WM_X))
y_train = np.vstack((y_train.reshape(-1, 1), WM_y)).ravel()
ownership['inputs'] = WM_X
ownership['labels'] = WM_y.ravel()
ownership['bounds'] = (min(y_train), max(y_train))

return ownership, X_train, y_train

def train_step(self, ownership, X_train, y_train):
"""Train the model depending on model type.

Args:
ownership (dict): Ownership information about
triggers
X_train (array): Input data
y_train (array): Label data (0 / 1)

Returns:
None for classification tasks