fixed a bug on the chronos pipeline

Author: ryuta-yoshimatsu

examples/foundation_daily.py

@@ -129,7 +129,6 @@ def transform_group(df):
     "MoiraiLarge",
     "MoiraiMoESmall",
     "MoiraiMoEBase",
-    "MoiraiMoELarge",
     "TimesFM_1_0_200m",
     "TimesFM_2_0_500m",
 ]

examples/foundation_monthly.py

@@ -135,7 +135,6 @@ def transform_group(df):
     "MoiraiLarge",
     "MoiraiMoESmall",
     "MoiraiMoEBase",
-    "MoiraiMoELarge",
     "TimesFM_1_0_200m",
     "TimesFM_2_0_500m",
 ]

examples/m5-examples/foundation_daily_m5.py

@@ -35,7 +35,6 @@
     "MoiraiLarge",
     "MoiraiMoESmall",
     "MoiraiMoEBase",
-    "MoiraiMoELarge",
     "TimesFM_1_0_200m",
     "TimesFM_2_0_500m",
 ]

mmf_sa/Forecaster.py

@@ -383,7 +383,6 @@ def backtest_global_model(
                 spark=self.spark,
                 # backtest_retrain=self.conf["backtest_retrain"],
             ))
-
         group_id_dtype = IntegerType() \
             if train_df[self.conf["group_id"]].dtype == 'int' else StringType()
 
@@ -399,7 +398,6 @@ def backtest_global_model(
             ]
         )
         res_sdf = self.spark.createDataFrame(res_pdf, schema)
-
         # Write evaluation results to a delta table
         if write:
             if self.conf.get("evaluation_output", None):
@@ -413,7 +411,6 @@ def backtest_global_model(
                     .write.mode("append")
                     .saveAsTable(self.conf.get("evaluation_output"))
                 )
-
         # Compute aggregated metrics
         res_df = (
             res_sdf.groupby(["metric_name"])

mmf_sa/models/chronosforecast/ChronosPipeline.py

@@ -90,15 +90,13 @@ def predict(self,
                 horizon_timestamps_udf(hist_df.ds).alias("ds"),
                 forecast_udf(hist_df.y).alias("y"))
         ).toPandas()
-
         forecast_df = forecast_df.reset_index(drop=False).rename(
             columns={
                 "unique_id": self.params.group_id,
                 "ds": self.params.date_col,
                 "y": self.params.target,
             }
         )
-
         # Todo
         # forecast_df[self.params.target] = forecast_df[self.params.target].clip(0.01)
         return forecast_df, self.model
@@ -165,19 +163,13 @@ def predict_udf(bulk_iterator: Iterator[pd.Series]) -> Iterator[pd.Series]:
             import numpy as np
             import pandas as pd
             # Initialize the ChronosPipeline with a pretrained model from the specified repository
-            from chronos import BaseChronosPipeline, ChronosBoltPipeline
-            if "bolt" in self.repo:
-                pipeline = ChronosBoltPipeline.from_pretrained(
-                    self.repo,
-                    device_map=self.device,
-                    torch_dtype=torch.bfloat16,
-                )
-            else:
-                pipeline = BaseChronosPipeline.from_pretrained(
-                    self.repo,
-                    device_map=self.device,
-                    torch_dtype=torch.bfloat16,
-                )
+            from chronos import BaseChronosPipeline
+            pipeline = BaseChronosPipeline.from_pretrained(
+                self.repo,
+                device_map='cuda',
+                torch_dtype=torch.bfloat16,
+            )
+
             # inference
             for bulk in bulk_iterator:
                 median = []
@@ -262,19 +254,12 @@ def __init__(self, repository, prediction_length):
         self.prediction_length = prediction_length
         self.device = "cuda" if torch.cuda.is_available() else "cpu"
         # Initialize the ChronosPipeline with a pretrained model from the specified repository
-        from chronos import BaseChronosPipeline, ChronosBoltPipeline
-        if "bolt" in self.repository:
-            self.pipeline = ChronosBoltPipeline.from_pretrained(
-                self.repository,
-                device_map=self.device,
-                torch_dtype=torch.bfloat16,
-            )
-        else:
-            self.pipeline = BaseChronosPipeline.from_pretrained(
-                self.repository,
-                device_map=self.device,
-                torch_dtype=torch.bfloat16,
-            )
+        from chronos import BaseChronosPipeline
+        self.pipeline = BaseChronosPipeline.from_pretrained(
+            self.repository,
+            device_map='cuda',
+            torch_dtype=torch.bfloat16,
+        )
 
     def predict(self, context, input_data, params=None):
         history = [torch.tensor(list(series)) for series in input_data]