ValueError: Could not infer frequency for series

[francoisabcd]

Hi,

I have a dataset that contains data only for working days. Since weekends and holidays are missing, I’m getting the following error:

ValueError: Could not infer frequency for series

Thank you,
Regards

[shchur]

Hi @francoisabcd, the recommended approach is to convert the dataset to daily frequency with tsdf.convert_frequency("D") (https://auto.gluon.ai/stable/api/autogluon.timeseries.TimeSeriesDataFrame.convert_frequency.html). This will put missing values NaN for the timestamps with missing observations that the TimeSeriesPredictor can natively handle.

[shchur]

Also see https://auto.gluon.ai/stable/tutorials/timeseries/forecasting-indepth.html#handling-irregular-data-and-missing-values

[francoisabcd]

Hi,

df = df.resample("D").last() can be used to fill in missing days with NaN values in a DataFrame.

Do I need to explicitly specify the past covariates in df (as context)? Or will the model automatically treat all columns in df as past covariates if I don’t provide a future_df (i.e., no future covariates)?

Thank you,

[shchur]

If you don't provide the future df, all columns are interpreted as past covariates

[francoisabcd]

I’m curious about how the model handles weekends within its architecture, given that it generates forecast values for those days. What role does reindexing the DataFrame play in this process—does the model actually learn meaningful patterns from the date indices?

During fine-tuning, it appears that all-time series are provided together as input. How does the model determine the appropriate context window in this case? Would it be more effective to segment the time series into fixed-length windows and provide multiple smaller DataFrames instead?

Thank you,