Merge branch 'main' into joshr17-patch-1

examples/lightgbm_node.py

@@ -25,9 +25,6 @@
 parser.add_argument("--dataset", type=str, default="rel-stack")
 parser.add_argument("--task", type=str, default="user-engage")
 parser.add_argument("--num_trials", type=int, default=10)
-# Use auto-regressive label as hand-crafted feature as input to LightGBM
-parser.add_argument("--use_ar_label", action="store_true", default=False)
-parser.add_argument("--no-use_ar_label", dest="use_ar_label", action="store_false")
 parser.add_argument(
     "--sample_size",
     type=int,
@@ -54,46 +51,6 @@
 val_table = task.get_table("val")
 test_table = task.get_table("test")
 
-ar_label_cols = []
-
-if args.use_ar_label:
-    ### Adding AR labels into train/val/test_table
-    whole_df = pd.concat([train_table.df, val_table.df, test_table.df], axis=0)
-    num_ar_labels = max(train_table.df[train_table.time_col].nunique() - 2, 1)
-
-    sorted_unique_times = np.sort(whole_df[train_table.time_col].unique())
-    timedelta = sorted_unique_times[1:] - sorted_unique_times[:-1]
-    if (timedelta / timedelta[0] - 1).max() > 0.1:
-        raise RuntimeError(
-            "Timestamps are not equally spaced, making it inappropriate for "
-            "AR labels to be used."
-        )
-    TIME_IDX_COL = "time_idx"
-    time_df = pd.DataFrame(
-        {
-            task.time_col: sorted_unique_times,
-            "time_idx": np.arange(len(sorted_unique_times)),
-        }
-    )
-
-    whole_df = whole_df.merge(time_df, how="left", on=task.time_col)
-    whole_df.drop(task.time_col, axis=1, inplace=True)
-    # Shift timestamp of whole_df iteratively and join it with train/val/test_table
-    for i in range(1, num_ar_labels + 1):
-        whole_df_shifted = whole_df.copy(deep=True)
-        # Shift time index by i
-        whole_df_shifted[TIME_IDX_COL] += i
-        # Map time index back to datetime timestamp
-        whole_df_shifted = whole_df_shifted.merge(time_df, how="inner", on=TIME_IDX_COL)
-        whole_df_shifted.drop(TIME_IDX_COL, axis=1, inplace=True)
-        ar_label = f"AR_{i}"
-        ar_label_cols.append(ar_label)
-        whole_df_shifted.rename(columns={task.target_col: ar_label}, inplace=True)
-
-        for table in [train_table, val_table, test_table]:
-            table.df = table.df.merge(
-                whole_df_shifted, how="left", on=(task.entity_col, task.time_col)
-            )
 
 dfs: Dict[str, pd.DataFrame] = {}
 entity_table = dataset.get_db().table_dict[task.entity_table]
@@ -117,16 +74,10 @@
 
 if task.task_type == TaskType.BINARY_CLASSIFICATION:
     col_to_stype[task.target_col] = torch_frame.categorical
-    for ar_label in ar_label_cols:
-        col_to_stype[ar_label] = torch_frame.categorical
 elif task.task_type == TaskType.REGRESSION:
     col_to_stype[task.target_col] = torch_frame.numerical
-    for ar_label in ar_label_cols:
-        col_to_stype[ar_label] = torch_frame.numerical
 elif task.task_type == TaskType.MULTILABEL_CLASSIFICATION:
     col_to_stype[task.target_col] = torch_frame.embedding
-    for ar_label in ar_label_cols:
-        col_to_stype[ar_label] = torch_frame.embedding
 else:
     raise ValueError(f"Unsupported task type called {task.task_type}")