search

AIStream-Peelout / flow-forecast

Deep learning PyTorch library for time series forecasting, classification, and anomaly detection (originally for flood forecasting).
https://flow-forecast.atlassian.net/wiki/spaces/FF/overview
GNU General Public License v3.0
2.06k stars 293 forks source link

Double check test_loss #253

Open isaacmg opened 3 years ago

isaacmg commented 3 years ago

Test loss seems suspiciously low in a number of recent experiments.

isaacmg commented 3 years ago

It looks like DilateLoss is correlated with MSE when forecast_length == 1, however the tensor might be in the improper dimension when it is passed to dilate loss. In any case, the problem appears to be with DilateLoss and does not affect other loss functions. We will continue to update this issue rest assured any results reported in terms of MSE or loss functions other than DilateLoss are accurate. Remember to include forecast_test_len if you want the test loss to be over a longer period.