Spaces:
Sleeping
Sleeping
File size: 3,585 Bytes
9d61c9b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 |
from typing import List, Union
import numpy as np
# DEPRECATED: for dataset preprocessing use torch version.
# Here is datatype issue this code converts torch.float32 to numpy.float64
# And it causes error in the model training:
# Error: RuntimeError: Input type (torch.cuda.DoubleTensor) and weight type (torch.cuda.FloatTensor) should be the same!
def pad_1D(inputs: List[np.ndarray], pad_value: float = 0.0) -> np.ndarray:
r"""Pad a list of 1D numpy arrays to the same length.
Args:
inputs (List[np.ndarray]): List of 1D numpy arrays to pad.
pad_value (float): Value to use for padding. Default is 0.0.
Returns:
np.ndarray: Padded 2D numpy array of shape (len(inputs), max_len), where max_len is the length of the longest input array.
"""
def pad_data(x: np.ndarray, length: int) -> np.ndarray:
r"""Pad a 1D numpy array with zeros to a specified length.
Args:
x (np.ndarray): 1D numpy array to pad.
length (int): Length to pad the array to.
Returns:
np.ndarray: Padded 1D numpy array of shape (length,).
"""
return np.pad(
x, (0, length - x.shape[0]), mode="constant", constant_values=pad_value,
)
max_len = max(len(x) for x in inputs)
return np.stack([pad_data(x, max_len) for x in inputs])
def pad_2D(
inputs: List[np.ndarray], maxlen: Union[int, None] = None, pad_value: float = 0.0,
) -> np.ndarray:
r"""Pad a list of 2D numpy arrays to the same length.
Args:
inputs (List[np.ndarray]): List of 2D numpy arrays to pad.
maxlen (Union[int, None]): Maximum length to pad the arrays to. If None, pad to the length of the longest array. Default is None.
pad_value (float): Value to use for padding. Default is 0.0.
Returns:
np.ndarray: Padded 3D numpy array of shape (len(inputs), max_len, input_dim), where max_len is the maximum length of the input arrays, and input_dim is the dimension of the input arrays.
"""
def pad(x: np.ndarray, max_len: int) -> np.ndarray:
r"""Pad a 2D numpy array with zeros to a specified length.
Args:
x (np.ndarray): 2D numpy array to pad.
max_len (int): Maximum length to pad the array to.
Returns:
np.ndarray: Padded 2D numpy array of shape (x.shape[0], max_len), where x.shape[0] is the number of rows in the input array.
"""
if np.shape(x)[1] > max_len:
raise ValueError("not max_len")
padding = np.ones((x.shape[0], max_len - np.shape(x)[1])) * pad_value
return np.concatenate((x, padding), 1)
if maxlen:
output = np.stack([pad(x, maxlen) for x in inputs])
else:
max_len = max(np.shape(x)[1] for x in inputs)
output = np.stack([pad(x, max_len) for x in inputs])
return output
def pad_3D(inputs: Union[np.ndarray, List[np.ndarray]], B: int, T: int, L: int) -> np.ndarray:
r"""Pad a 3D numpy array to a specified shape.
Args:
inputs (np.ndarray): 3D numpy array to pad.
B (int): Batch size to pad the array to.
T (int): Time steps to pad the array to.
L (int): Length to pad the array to.
Returns:
np.ndarray: Padded 3D numpy array of shape (B, T, L), where B is the batch size, T is the time steps, and L is the length.
"""
inputs_padded = np.zeros((B, T, L), dtype=np.float32)
for i, input_ in enumerate(inputs):
inputs_padded[i, : np.shape(input_)[0], : np.shape(input_)[1]] = input_
return inputs_padded
|