python - 信号开发/猎户座改变 TadGAN 的窗口大小

Question

我正在使用本文中引用的 Orion 对于 tadGAN 管道，我无法更改 window_size（用于调整模型），如下面的代码所示。这适用于这个包中的其他模型（带阈值的 LSTM 工作正常）

任何想法为什么？

（Google Colab 中的可重现示例）

! pip install orion-ml 'urllib3>=1.25.4,<1.26'

# to get tadgan.json 
%%bash
rm -rf Orion
rm -rf images

git clone https://github.com/signals-dev/Orion.git
mv Orion/notebooks/tulog/* .
exit

# Packages
from orion import Orion
from orion.data import load_signal
import shutil

# Move json into working directory
shutil.copy("/content/Orion/orion/pipelines/verified/tadgan/tadgan.json", "/content")
shutil.copy("/content/Orion/orion/pipelines/verified/lstm_dynamic_threshold/lstm_dynamic_threshold.json", "/content")

# Select data 
signal = 'nyc_taxi'

# load signal
df = load_signal(signal)

parameters = {
    "mlprimitives.custom.timeseries_preprocessing.time_segments_aggregate#1": {
            "interval": 3600 # hour level
        },
    "mlprimitives.custom.timeseries_preprocessing.rolling_window_sequences#1": {
            "target_column": 0,
            "window_size": 50
    }
}

# with tadGAN
orion = Orion(
    'tadgan.json',
    parameters
)

anomalies = orion.fit_detect(df)

给出这个错误：

Exception caught fitting MLBlock orion.primitives.tadgan.TadGAN#1
Traceback (most recent call last):
  File "/usr/local/lib/python3.7/dist-packages/mlblocks/mlpipeline.py", line 549, in _fit_block
    block.fit(**fit_args)
  File "/usr/local/lib/python3.7/dist-packages/mlblocks/mlblock.py", line 302, in fit
    getattr(self.instance, self.fit_method)(**fit_kwargs)
  File "/usr/local/lib/python3.7/dist-packages/orion/primitives/tadgan.py", line 251, in fit
    X = X.reshape((-1, self.shape[0], 1))
ValueError: cannot reshape array of size 255550 into shape (100,1)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-7-27445b4ee358> in <module>()
     30 )
     31 
---> 32 anomalies = orion.fit_detect(df)

5 frames
/usr/local/lib/python3.7/dist-packages/orion/primitives/tadgan.py in fit(self, X, **kwargs)
    249         """
    250         self._build_tadgan(**kwargs)
--> 251         X = X.reshape((-1, self.shape[0], 1))
    252         self._fit(X)
    253 

ValueError: cannot reshape array of size 255550 into shape (100,1)

但它适用于不同的模型（使用上面定义的参数）

# with LSTM

lstm_orion = Orion(
    'lstm_dynamic_threshold.json',
    parameters
)

lstm_anomalies = lstm_orion.fit_detect(df)
lstm_anomalies

给出这样的结果（如预期的那样！）

start   end severity
0   1404345600  1404694800  0.161157
1   1414706400  1415145600  0.655987
2   1422147600  1422590400  0.503789

如何更改 tadGAN 的窗口大小？

标签： pythontime-seriesoutliersgenerative-adversarial-networkanomaly-detection