Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

Hi team! Do we have a Databricks account or instance I can use? I have users from Unilever and Lam reporting errors using the client in Databricks with Spark dataframes, and it's really hard for me to debug and understand why it happens. I can't reproduce their errors in Colab and using Fugue.

By the way, this error was reported twice:
```RecursionError: maximum recursion depth exceeded
File &lt;command-6950378966334387&gt;, line 1
----&gt; 1 fcst = client.forecast(
      2     sdf,
      3     h=30
      4 )
File /local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.12/site-packages/nixtla/nixtla_client.py:1536, in NixtlaClient.forecast(self, df, h, freq, id_col, time_col, target_col, X_df, level, quantiles, finetune_steps, finetune_depth, finetune_loss, finetuned_model_id, clean_ex_first, hist_exog_list, validate_api_key, add_history, date_features, date_features_to_one_hot, model, num_partitions, feature_contributions)
   1450 """Forecast your time series using TimeGPT.
   1451 
   1452 Parameters
   (...)
   1533     predictions (if level is not None).
   1534 """
   1535 if not isinstance(df, (pd.DataFrame, pl_DataFrame)):
-&gt; 1536     return self._distributed_forecast(
   1537         df=df,
   1538         h=h,
   1539         freq=freq,
   1540         id_col=id_col,```
But I can't reproduce it in Colab

Tagging <@U066SSLDSPL> <@U08EZBX8UQH> to access Databricks and get help debugging.

I think a recursive call happens because of type checking the dataframe. This seems to only happen with the python wheel and nixtla=<https://github.com/Nixtla/nixtla/releases/tag/v0.6.7.dev3|v0.6.7.dev3>.

I can't reproduce the error with the dev API in Colab and using Fugue (like in our tutorial).

We will take a look in a couple of hours. So tomorrow when you wake up it should be done. 

I figured out the problem, there will be a small fix on the large model repo, we are lucky because we don't have to change the client code