代码拉取完成，页面将自动刷新
Also preprocessors acting on numpy arrays.
Preprocessors do the job of ECG signal preprocessing before fed into some neural network and are monitored by a manager. A short example is as follows
import torch
from torch_ecg.cfg import CFG
from torch_ecg.preprocessors import PreprocManager
config = CFG(
random=False,
bandpass={"fs":500},
normalize={"method": "min-max"},
)
ppm = PreprocManager.from_config(config)
sig = torch.randn(2, 12, 8000)
sig = ppm(sig)
For Preprocessors that operate on the numpy arrays, see the following example
import torch
from torch_ecg.cfg import CFG
from torch_ecg._preprocessors import PreprocManager
config = CFG(
random=False,
resample={"fs": 500},
bandpass={"filter_type": "fir"},
normalize={"method": "min-max"},
)
ppm = PreprocManager.from_config(config)
sig = torch.randn(12, 80000).numpy()
sig, fs = ppm(sig, 200)
One can create custom preprocessors to be maintained by the manager. The following is a simple example
import torch
from torch_ecg.cfg import CFG
from torch_ecg.preprocessors import PreprocManager
class DummyPreProcessor(torch.nn.Module):
"""
a dummy preprocessor that does nothing, similar to `torch.nn.Identity`
"""
__name__ = "DummyPreProcessor"
def forward(self, sig:torch.Tensor) -> torch.Tensor:
"""
"""
return sig
config = CFG(
random=False,
bandpass={"fs":500},
normalize={"method": "min-max"},
)
ppm = PreprocManager.from_config(config)
dp = DummyPreProcessor()
ppm.add_(dp, pos=1)
sig = torch.randn(2, 12, 8000)
sig = ppm(sig)
Here is another example for numpy version custom preprocessors
from numbers import Real
from typing import Tuple
import numpy as np
import torch
from torch_ecg.cfg import CFG
from torch_ecg._preprocessors import PreprocManager, PreProcessor
class DummyPreProcessor(PreProcessor):
"""
a dummy preprocessor that does nothing
"""
__name__ = "DummyPreProcessor"
def apply(self, sig:np.ndarray, fs:Real) -> Tuple[np.ndarray, int]:
"""
"""
return sig, fs
config = CFG(
random=False,
resample={"fs": 500},
bandpass={"filter_type": "fir"},
normalize={"method": "min-max"},
)
ppm = PreprocManager.from_config(config)
dp = DummyPreProcessor()
ppm.add_(dp, pos=1)
sig = torch.randn(12, 80000).numpy()
sig, fs = ppm(sig, 200)
The following preprocessors are implemented
Also known as detrending, via median filter, which removes baseline drifts.
Normalization methods including z-score normalization, min-max normalization, etc. are implemented. Refer to wikipedia for more details.
This procedure is performed using finite impulse response (FIR) filters, Butterworth filters, etc., which removes noises of frequencies outside the given pass band.
Resampling to a fixed sampling rate is done using the interpolate function from torch.nn.functional or using resample_poly function from scipy.signal.
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。