data.preprocs.ProcPad¶

Class · Source

proc = mdnc.data.preprocs.ProcPad(
    pad_width, inds=None, parent=None, **kwargs
)

This is a homogeneous processor. Use np.pad to pad the data.

This processor supports all np.pad options. Besides, this processor also support cropping. If any element in the argument pad_width is negative, would perform cropping on that axis. For example:

p = ProcPad(pad_width=((5, -5),))
y = p(x)  # x.shape=(20,), y.shape=(20,)

In this case, the data is padded by 5 samples at the beginning, but cropped 5 samples at the end.

This operator is not invertible when cropping is applied. The postprocess would try to revert the padding / cropping configurations to match the input data.

Arguments¶

Requries

Methods¶

preprocess¶

y_1, y_2, ... = proc.preprocess(x_1, x_2, ...)

The preprocess function. Perform the padding / cropping.

If parent exists, the input of this function comes from the output of parent.preprocess(). Otherwise, the input would comes from the input varibable directly.

Requries

Returns

postprocess¶

x_1, x_2, ... = proc.postprocess(y_1, y_2, ...)

The postprocess function. The inverse operator is not invertible if cropping is used in preprocess(). In this case, the cropped part would be padded by processed data (y, ).

If parent exists, the output of this function would be passed as the input of parent.postprocess(). Otherwise, the output would be returned to users directly.

Requries

Returns

Properties¶

pad_width¶

proc.pad_width

The padding width of the processor.

parent¶

proc.parent

The parent processor of this instance. The processor is also a derived class of ProcAbstract. If the parent does not exist, would return None.

has_ind¶

proc.has_ind

A bool flag, showing whether this processor and its all parent processors have inds configured or initialized with _disable_inds. In this case, the arguments of preprocess() and postprocess() would not share the same operation. We call such kind of processors "Inhomogeneous processors".

Examples¶

The processor need to be derived. We have two ways to implement the derivation, see the following examples.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20

import numpy as np
import matplotlib.pyplot as plt
import mdnc

proc = mdnc.data.preprocs.ProcPad(pad_width=((0, 0), (10, -10)), mode='wrap')
random_rng = np.random.default_rng()
data = random_rng.uniform(low=0.0, high=1.0, size=[10, 30])
t = proc.preprocess(data)
t_b = proc.postprocess(t)

with mdnc.utils.draw.setFigure(font_size=12):
    fig, axs = plt.subplots(nrows=3, ncols=1, sharex=True, figsize=(12, 5))
    axs[0].plot(t[0])
    axs[1].plot(t_b[0])
    axs[2].plot(data[0])
    axs[0].set_ylabel('Preprocessing')
    axs[1].set_ylabel('Inversed\npreprocessing')
    axs[2].set_ylabel('Raw\ndata')
    plt.tight_layout()
    plt.show()

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

import numpy as np
import matplotlib.pyplot as plt
import mdnc

proc = mdnc.data.preprocs.ProcPad(pad_width=((0, 0), (10, -10), (-10, 10)), mode='constant', constant_values=0.0)
random_rng = np.random.default_rng()
data = random_rng.uniform(low=0.0, high=1.0, size=[10, 30, 30])
t = proc.preprocess(data)
t_b = proc.postprocess(t)

with mdnc.utils.draw.setFigure(font_size=12):
    fig, axs = plt.subplots(nrows=1, ncols=3, sharex=True, sharey=True, figsize=(12, 4))
    im1 = axs[0].imshow(t[2])
    im2 = axs[1].imshow(t_b[0])
    im3 = axs[2].imshow(data[0])
    fig.colorbar(im1, ax=axs[0], pad=0.1, orientation='horizontal')
    fig.colorbar(im2, ax=axs[1], pad=0.1, orientation='horizontal')
    fig.colorbar(im3, ax=axs[2], pad=0.1, orientation='horizontal')
    axs[0].set_ylabel('Preprocessing')
    axs[1].set_ylabel('Inversed\npreprocessing')
    axs[2].set_ylabel('Raw data')
    plt.tight_layout()
    plt.show()

Comments