search

timeseriesAI / tsai

Time series Timeseries Deep Learning Machine Learning Python Pytorch fastai | State-of-the-art Deep Learning library for Time Series and Sequences in Pytorch / fastai
https://timeseriesai.github.io/tsai/
Apache License 2.0
5.07k stars 633 forks source link

ResNetPlus, FCNPlus, InceptionRocketPlus, MiniRocketPlus and XResNet1dPlus don't work with ndim fcst targets #718

Closed oguiza closed 1 year ago

oguiza commented 1 year ago

All these architectures fail the following test: 

from tsai.basics import *
from tsai.learner import all_arch_names

X=torch.rand(8,2,50)
y=torch.rand(8,1)
splits = TimeSplitter(show_plot=False)(y)

fail_test = []
for arch in all_arch_names:
    if not "plus" in arch.lower(): continue
    try:
        fcst = TSForecaster(X, y, splits=splits, arch=arch, metrics=mse)
        with ContextManagers([fcst.no_bar(), fcst.no_logging()]):
            fcst.fit_one_cycle(1, 1e-3)
    except Exception as e: 
        fail_test.append(arch)
        print(arch, e)

test_eq(fail_test, [])

Need to confirm if it's possible to update them so that they can be used with multidimensional targets.