class name constructed from task_type; cleaned unused pieces

presto/eval.py

@@ -9,57 +9,46 @@
 import pandas as pd
 import torch
 from catboost import CatBoostClassifier, Pool
-
-from .hierarchical_classification import (
-    CatBoostClassifierWrapper,
-    LocalClassifierPerNodeWrapper,
-    LocalClassifierPerParentNodeWrapper,
-)
-from hiclass import LocalClassifierPerParentNode, LocalClassifierPerNode
-
+from hiclass import LocalClassifierPerNode, LocalClassifierPerParentNode
 from sklearn.base import BaseEstimator, clone
 from sklearn.ensemble import RandomForestClassifier
 from sklearn.linear_model import LogisticRegression
-from sklearn.metrics import f1_score, precision_score, recall_score, classification_report
+from sklearn.metrics import classification_report, f1_score, precision_score, recall_score
 from torch import nn
 from torch.optim import AdamW, lr_scheduler
 from torch.utils.data import DataLoader, TensorDataset
 from tqdm import tqdm
 
 from .dataset import (
-    NORMED_BANDS,
     CLASS_MAPPINGS,
+    NORMED_BANDS,
     WorldCerealInferenceDataset,
     WorldCerealLabelled10DDataset,
     WorldCerealLabelledDataset,
 )
-from .presto import (
-    Presto,
-    PrestoFineTuningModel,
-    get_sinusoid_encoding_table,
-    param_groups_lrd,
+from .hierarchical_classification import (
+    CatBoostClassifierWrapper,
+    LocalClassifierPerNodeWrapper,
+    LocalClassifierPerParentNodeWrapper,
 )
+from .presto import Presto, PrestoFineTuningModel, get_sinusoid_encoding_table, param_groups_lrd
 from .utils import DEFAULT_SEED, device
 
-
 logger = logging.getLogger("__main__")
 
 SklearnStyleModel = Union[BaseEstimator, CatBoostClassifier]
 
 
 @dataclass
 class Hyperparams:
-    # lr: float = 2e-5
-    lr: float = 0.0001
+    lr: float = 2e-5
     max_epochs: int = 1000
     batch_size: int = 256
-    # batch_size: int = 2048
     patience: int = 20
     num_workers: int = 8
 
 
 class WorldCerealEval:
-    name = "WorldCerealCroptype"
     threshold = 0.5
     regression = False
 
@@ -82,11 +71,9 @@ def __init__(
         finetune_classes: str = "CROPTYPE0",
     ):
         self.seed = seed
-
-        if name is not None:
-            self.name = name
         self.target_function = target_function
         self.task_type = task_type
+        self.name = f"WorldCereal{task_type.title()}"
 
         train_data, val_data = WorldCerealLabelledDataset.split_df(train_data, val_size=val_size)
         self.train_df = self.prep_dataframe(train_data, filter_function, dekadal=dekadal)
@@ -200,8 +187,6 @@ def _construct_finetuning_model(self, pretrained_model: Presto) -> PrestoFineTun
     @torch.no_grad()
     def finetune_sklearn_model(
         self,
-        # dl: DataLoader,
-        # val_dl: DataLoader,
         pretrained_model: PrestoFineTuningModel,
         models: List[str] = ["Regression", "Random Forest"],
     ) -> Union[Sequence[BaseEstimator], Dict]:
@@ -406,7 +391,7 @@ def _inference_for_dl(
                     probs = preds.copy()
                 if task_type == "croptype":
                     preds = finetuned_model.predict(encodings)
-                    
+
                     # for hierarchical classification, get predictions on the most granular level
                     if preds.ndim > 1:
                         preds = preds[:, -1]
@@ -454,14 +439,14 @@ def spatial_inference(
             if self.task_type == "croptype":
                 if pretrained_model is None:
                     temp_croptype_map = pd.DataFrame(
-                        CLASS_MAPPINGS["CROPTYPE0"].items(), columns=["ewoc_code", "name"]
+                        CLASS_MAPPINGS["CROPTYPE19"].items(), columns=["ewoc_code", "name"]
                     )
                     test_preds_np = np.argmax(test_preds_np, axis=-1)
                     test_preds_str = np.array([self.croptype_list[xx] for xx in test_preds_np])
                 else:
                     test_preds_np = np.argmax(test_probs_np, axis=-1)
                     temp_croptype_map = pd.DataFrame(
-                        CLASS_MAPPINGS["CROPTYPE9"].items(), columns=["ewoc_code", "name"]
+                        CLASS_MAPPINGS["CROPTYPE19"].items(), columns=["ewoc_code", "name"]
                     )
 
                 temp_croptype_map.sort_values(
@@ -486,9 +471,17 @@ def spatial_inference(
                 df = ds.combine_predictions(
                     latlons, test_preds_np, test_preds_np, test_preds_np, y, ndvi, b2, b3, b4
                 )
-            if self.task_type == "croptype":    
+            if self.task_type == "croptype":
                 df = ds.combine_predictions(
-                    latlons, test_preds_np, test_preds_ewoc_code, test_probs_np, y, ndvi, b2, b3, b4
+                    latlons,
+                    test_preds_np,
+                    test_preds_ewoc_code,
+                    test_probs_np,
+                    y,
+                    ndvi,
+                    b2,
+                    b3,
+                    b4,
                 )
             prefix = f"{self.name}_{ds.all_files[i].stem}"
             if pretrained_model is None:
@@ -778,35 +771,7 @@ def finetuning_results_sklearn(
     ):
         results_df = pd.DataFrame()
         if len(sklearn_model_modes) > 0:
-            # dl = DataLoader(
-            #     WorldCerealLabelledDataset(
-            #         self.train_df,
-            #         countries_to_remove=self.countries_to_remove,
-            #         years_to_remove=self.years_to_remove,
-            #         target_function=self.target_function,
-            #         task_type=self.task_type,
-            #         croptype_list=[],
-            #     ),
-            #     batch_size=2048,
-            #     shuffle=False,
-            #     num_workers=8,
-            # )
-            # val_dl = DataLoader(
-            #     WorldCerealLabelledDataset(
-            #         self.val_df,
-            #         countries_to_remove=self.countries_to_remove,
-            #         years_to_remove=self.years_to_remove,
-            #         target_function=self.target_function,
-            #         task_type=self.task_type,
-            #         croptype_list=[],
-            #     ),
-            #     batch_size=2048,
-            #     shuffle=False,
-            #     num_workers=8,
-            # )
             sklearn_models = self.finetune_sklearn_model(
-                # dl,
-                # val_dl,
                 finetuned_model,
                 models=sklearn_model_modes,
             )