测试集长度问题 #561
Description
import numpy as np
from paddlets import TSDataset
from paddlets.models.forecasting import RNNBlockRegressor, MLPRegressor, LSTNetRegressor
import matplotlib.pyplot as plt
data = TSDataset.load_from_csv("demo.csv", target_cols=['Q'], time_col='datetime',
fillna_method="pre", fill_missing_dates=True)
model = LSTNetRegressor(in_chunk_len=30, out_chunk_len=3)
train, val_test = data.split(0.8)
val, test = val_test.split(0.5)
model.fit(train_tsdataset=train, valid_tsdataset=val)
predicted_dataset = model.predict(test)D:\AnacondaEnvs\energy-prediction\python.exe D:\02sourcecode\PycharmProjects\energy-prediction\demo.py
[2025-08-15 09:53:33,557] [paddlets] [ERROR] ValueError: if time_window[1] (5) <= len(TSDataset.target) - 1 (5), then 32 <= time_window[0] (32) <= time_window[1] (5) <= len(TSDataset.target) - 1 (5) must be True.
Traceback (most recent call last):
File "D:\02sourcecode\PycharmProjects\energy-prediction\demo.py", line 14, in <module>
model.fit(train_tsdataset=train, valid_tsdataset=val)
File "C:\Users\YPRJ\AppData\Roaming\Python\Python39\site-packages\paddlets\models\forecasting\dl\paddle_base_impl.py", line 345, in fit
train_dataloader, valid_dataloaders = self._init_fit_dataloaders(train_tsdataset, valid_tsdataset)
File "C:\Users\YPRJ\AppData\Roaming\Python\Python39\site-packages\paddlets\models\forecasting\dl\paddle_base_impl.py", line 225, in _init_fit_dataloaders
dataset = data_adapter.to_sample_dataset(
File "C:\Users\YPRJ\AppData\Roaming\Python\Python39\site-packages\paddlets\models\data_adapter.py", line 1507, in to_sample_dataset
return SampleDataset(
File "C:\Users\YPRJ\AppData\Roaming\Python\Python39\site-packages\paddlets\models\data_adapter.py", line 275, in __init__
self._validate_time_window()
File "C:\Users\YPRJ\AppData\Roaming\Python\Python39\site-packages\paddlets\models\data_adapter.py", line 925, in _validate_time_window
raise_if_not(
File "C:\Users\YPRJ\AppData\Roaming\Python\Python39\site-packages\paddlets\logger\logger.py", line 135, in raise_if_not
raise ValueError(message)
ValueError: if time_window[1] (5) <= len(TSDataset.target) - 1 (5), then 32 <= time_window[0] (32) <= time_window[1] (5) <= len(TSDataset.target) - 1 (5) must be True.
Process finished with exit code 1demo.csv 总共有61条数据。按照api文档说 in_chunk_len 是训练数据的长度，out_chunk_len 是预测产生的长度。那么我理解是用前30的数据，预测后面3条数据。
但是在训练的时候，我设置训练集、验证集、测试集。分别设置0.8和0.5，对应的分别是610.8， 610.20.5， 610.2*0.5 ，那么训练集、验证集、测试集分别是【48条,6条,6条】。 怎么会报这个错误。这个长度该怎么设置。
Metadata
Metadata
Assignees
Labels
No labels