Investigate if `convert_df_input_to_list_of_dicts_input` can be made faster for large datasets

[shchur]

Check if the method can be made more efficient for large datasets

chronos-forecasting/src/chronos/df_utils.py

Line 212 in 7daaa71

def convert_df_input_to_list_of_dicts_input(

[abdulfatir]

Adding some details here. For a dataframe with ~8K time series of length 200 each, here are the times.

• Input preprocessing time: 6.79s
  • DF Validation time: 1.47s
  • Input construction time: 5.31s
    • Target extraction time: 2.31s
    • Future timestamp creation time: 2.34s
• Prediction time: 2.67s
• Post-processing time: 2.84s

It's absurd that just extracting the target(s) from the dataframe and constructing the future index takes such a long time.