Tim Rosenflanz comments

Results 19 comments of


                                            Tim Rosenflanz

Multi-GPU training results in "ProcessExitedException process 0 terminated with signal SIGSEGV" exception for Baseline and TFT models.

I fixed mine by launching the code as a script and using`strategy="auto"` in the Trainer without any real changes to the model/dataset. Unfortunately, notebooks are notorious for having issues with...

Multi-GPU training results in "ProcessExitedException process 0 terminated with signal SIGSEGV" exception for Baseline and TFT models.

> Are you saying that this is all you changed? You didn't have to effectively rebuild the the TFT class? So fix would be >> run a script not a...

Do groups learn together or separately?

Your understanding is correct for the TemporalFusionTransformer model. You can also see that memory consumption doesn't really grow if more groups are added. Not sure about other models though

time_idx in TimeSeriesDataset should automatically be marked as time_varying_known_reals variable.

The error message is indeed hard to decode in the case when no time_varying values are provided. Note that there are some extra features that DataLoader can add like relative_time_idx...

DataFrameGroupBy.agg numeric_only incompatible with pd.NamedAgg

Having the same issue

DataFrameGroupBy.agg numeric_only incompatible with pd.NamedAgg

Found a workaround. Since the code only checks for `numeric_only = False`, we can just pass numeric_only as some random aggregation and it will pass as it is numeric_only is...

Is the ROC-AUC score calculation correct?

It might be a learning rate issue or something with the data prep since it is a bit more complicated than a simple LSTM. The ROC calculation should be correct....

Add KAN (Kolmogorov-Arnold Network) to Darts

Tried replacing nn.Linear with KANLinear from here https://github.com/Blealtan/efficient-kan in the TSMixer and it works quite well. The speed of iterations is slower (clock wise) as expected but fits much quicker...

How to mask or ignore target features during model training

I wish this issue was created 3 days ago when I wanted to do exactly the same. In my case I didn't want target and historic_future covariates to be included...

add `DataLoader` related parameters to `fit()` and `predict()`

> @madtoinou what do you think about hardcoded parameters like > > ```python > batch_size=self.batch_size, > shuffle=True, > drop_last=False, > collate_fn=self._batch_collate_fn, > ``` > > should it be hard coded...