Batch Normalization operator implemented in Numpy. Parameters ---------- data : np.ndarray Input to be batch-normalized. gamma : np.ndarray Scale factor to be applied to the normalized tensor. beta : np.ndarray Offset to be applied to the normalized ten
(
x: np.ndarray,
gamma: np.ndarray,
beta: np.ndarray,
moving_mean: np.ndarray,
moving_var: np.ndarray,
axis: int,
epsilon: float,
center: bool,
scale: bool,
training: bool,
momentum: float,
)
| 20 | |
| 21 | |
| 22 | def batch_norm( |
| 23 | x: np.ndarray, |
| 24 | gamma: np.ndarray, |
| 25 | beta: np.ndarray, |
| 26 | moving_mean: np.ndarray, |
| 27 | moving_var: np.ndarray, |
| 28 | axis: int, |
| 29 | epsilon: float, |
| 30 | center: bool, |
| 31 | scale: bool, |
| 32 | training: bool, |
| 33 | momentum: float, |
| 34 | ): |
| 35 | """Batch Normalization operator implemented in Numpy. |
| 36 | |
| 37 | Parameters |
| 38 | ---------- |
| 39 | data : np.ndarray |
| 40 | Input to be batch-normalized. |
| 41 | |
| 42 | gamma : np.ndarray |
| 43 | Scale factor to be applied to the normalized tensor. |
| 44 | |
| 45 | beta : np.ndarray |
| 46 | Offset to be applied to the normalized tensor. |
| 47 | |
| 48 | moving_mean : np.ndarray |
| 49 | Running mean of input. |
| 50 | |
| 51 | moving_var : np.ndarray |
| 52 | Running variance of input. |
| 53 | |
| 54 | axis : int |
| 55 | Specify along which shape axis the normalization should occur. |
| 56 | |
| 57 | epsilon : float |
| 58 | Small float added to variance to avoid dividing by zero. |
| 59 | |
| 60 | center : bool |
| 61 | If True, add offset of beta to normalized tensor, If False, |
| 62 | beta is ignored. |
| 63 | |
| 64 | scale : bool |
| 65 | If True, scale normalized tensor by gamma. If False, gamma |
| 66 | is ignored. |
| 67 | |
| 68 | training : bool |
| 69 | Indicating whether it is in training mode. If True, update |
| 70 | moving_mean and moving_var. |
| 71 | |
| 72 | momentum : float |
| 73 | The value used for the moving_mean and moving_var update |
| 74 | |
| 75 | Returns |
| 76 | ------- |
| 77 | output : np.ndarray |
| 78 | Normalized data with same shape as input |
| 79 |