Update prediction_length_map

Currently, the prediction length map maps the frequency of the data sets to their expected forecast horizon.

However, it uses pandas frequency strings which are outdated and causes issues when data loading.

This change simply adds new keys reflecting this change in pandas. There may be more keys that need to be updated but these are all I found.

monash_tsf.py

@@ -473,11 +473,15 @@ class MonashTSF(datasets.GeneratorBasedBuilder):
         if forecast_horizon is None:
             prediction_length_map = {
                 "S": 60,
+                "s": 60,
                 "T": 60,
+                "min": 60,
                 "H": 48,
+                "h": 48,
                 "D": 30,
                 "W": 8,
                 "M": 12,
+                "ME": 12,
                 "Y": 4,
             }
             freq = frequency_converter(frequency)