upgini 1.2.121a3__tar.gz → 1.2.122a2__tar.gz

{upgini-1.2.121a3 → upgini-1.2.122a2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: upgini
-Version: 1.2.121a3
+Version: 1.2.122a2
 Summary: Intelligent data search & enrichment for Machine Learning
 Project-URL: Bug Reports, https://github.com/upgini/upgini/issues
 Project-URL: Homepage, https://upgini.com/

upgini-1.2.122a2/src/upgini/__about__.py

@@ -0,0 +1 @@
+__version__ = "1.2.122a2"

{upgini-1.2.121a3 → upgini-1.2.122a2}/src/upgini/autofe/feature.py

@@ -42,6 +42,9 @@ class Column:
     def get_display_name(self, cache: bool = True, shorten: bool = False, **kwargs) -> str:
         return self.get_columns(**kwargs)[0]
 
+    def reset_display_indices(self) -> "Column":
+        return self
+
     def _unhash(self, feature_name: str) -> str:
         last_component_idx = feature_name.rfind("_")
         if not feature_name.startswith("f_"):
@@ -142,6 +145,7 @@ class Feature:
         )
 
     def rename_columns(self, mapping: Dict[str, str]) -> "Feature":
+        self.op.rename_columns(mapping)
         for child in self.children:
             child.rename_columns(mapping)
         self.cached_display_name = None
@@ -212,6 +216,13 @@ class Feature:
         self.cached_display_name = None
         return self
 
+    def reset_display_indices(self) -> "Feature":
+        for child in self.children:
+            child.reset_display_indices()
+        self.display_index = None
+        self.cached_display_name = None
+        return self
+
     def infer_type(self, data: pd.DataFrame) -> Union[str, DtypeObj]:
         if self.op.output_type:
             return self.op.output_type

{upgini-1.2.121a3 → upgini-1.2.122a2}/src/upgini/autofe/operator.py

@@ -89,6 +89,32 @@ class Operator(BaseModel, metaclass=OperatorRegistry):
     def delete_data(self):
         pass
 
+    def rename_columns(self, columns_renaming: Dict[str, str]) -> "Operator":
+        # Rename occurrences of column names inside self.params keys according to columns_renaming
+        if not self.params or not columns_renaming:
+            return self
+
+        # Replace longer keys first to avoid partial overlaps
+        replacements = sorted(columns_renaming.items(), key=lambda kv: -len(kv[0]))
+
+        renamed_params: Dict[str, str] = {}
+        for param_key, param_value in self.params.items():
+            new_key = param_key
+            for old, new in replacements:
+                if old and old in new_key:
+                    new_key = new_key.replace(old, new)
+
+            if new_key in renamed_params and new_key != param_key:
+                self._logger.warning(
+                    "Param key collision after rename: '%s' -> '%s'. Overwriting value.",
+                    param_key,
+                    new_key,
+                )
+            renamed_params[new_key] = param_value
+
+        self.params = renamed_params
+        return self
+
 
 class ParametrizedOperator(Operator, abc.ABC):
 

{upgini-1.2.121a3 → upgini-1.2.122a2}/src/upgini/features_enricher.py

@@ -1028,7 +1028,7 @@ class FeaturesEnricher(TransformerMixin):
                     columns_renaming,
                     _,
                 ) = prepared_data
-                
+
                 gc.collect()
 
                 if fitting_X.shape[1] == 0 and fitting_enriched_X.shape[1] == 0:
@@ -1406,7 +1406,7 @@ class FeaturesEnricher(TransformerMixin):
         self,
         X: pd.DataFrame,
         eval_set: list[tuple[pd.DataFrame, pd.Series]],
-        enriched_eval_set: dict,
+        enriched_eval_set: dict[int, tuple[pd.DataFrame, pd.Series, pd.DataFrame, pd.Series]],
         eval_set_dates: dict[int, pd.Series],
         search_keys: dict[str, SearchKey],
         stability_threshold: float,
@@ -1417,31 +1417,42 @@ class FeaturesEnricher(TransformerMixin):
         # Find latest eval set or earliest if all eval sets are before train set
         date_column = self._get_date_column(search_keys)
 
+        date_converter = DateTimeSearchKeyConverter(
+            date_column, self.date_format, self.logger, self.bundle, generate_cyclical_features=False
+        )
+
+        X = date_converter.convert(X)
+
         x_date = X[date_column].dropna()
-        if not is_numeric_dtype(x_date):
-            x_date = pd.to_datetime(x_date).dt.floor("D").astype(np.int64) / 10**6
-        main_min_date = x_date.min()
+        if len(x_date) == 0:
+            self.logger.warning("Empty date column in X")
+            return []
 
-        for eval_x, _ in eval_set:
-            eval_x_date = eval_x[date_column].dropna()
-            if not is_numeric_dtype(eval_x_date):
-                eval_x[date_column] = pd.to_datetime(eval_x_date).dt.floor("D").astype(np.int64) / 10**6
+        main_min_date = x_date.min()
 
         # Find minimum date for each eval_set and compare with main dataset
         eval_dates = []
         for i, (eval_x, _) in enumerate(eval_set):
-            if date_column in eval_x.columns:
-                if len(eval_x) < 1000:
-                    self.logger.warning(f"Eval_set {i} has less than 1000 rows. It will be ignored for stability check")
-                    continue
-                eval_x_date = eval_x[date_column].dropna()
-                if not is_numeric_dtype(eval_x_date):
-                    eval_x_date = pd.to_datetime(eval_x_date).dt.floor("D").astype(np.int64) / 10**6
-                eval_min_date = eval_x_date.min()
-                eval_max_date = eval_x_date.max()
-                eval_dates.append((i, eval_min_date, eval_max_date))
+            if date_column not in eval_x.columns:
+                self.logger.warning(f"Date column not found in eval_set {i + 1}")
+                continue
+            eval_x = date_converter.convert(eval_x)
+            eval_x_date = eval_x[date_column].dropna()
+            if len(eval_x_date) < 1000:
+                self.logger.warning(f"Eval_set {i} has less than 1000 rows. It will be ignored for stability check")
+                continue
+            if len(enriched_eval_set[i][2]) < 1000:
+                self.logger.warning(
+                    f"Enriched eval_set {i} has less than 1000 rows. It will be ignored for stability check"
+                )
+                continue
+
+            eval_min_date = eval_x_date.min()
+            eval_max_date = eval_x_date.max()
+            eval_dates.append((i, eval_min_date, eval_max_date))
 
         if not eval_dates:
+            self.logger.warning("There are no correct eval_sets for stability check")
             return []
 
         # Check if any eval_set has minimum date >= main dataset minimum date
@@ -1464,10 +1475,7 @@ class FeaturesEnricher(TransformerMixin):
         checking_eval_set_df = checking_eval_set_df.copy()
 
         checking_eval_set_df[date_column] = eval_set_dates[selected_eval_set_idx]
-        if not is_numeric_dtype(checking_eval_set_df[date_column]):
-            checking_eval_set_df[date_column] = (
-                pd.to_datetime(checking_eval_set_df[date_column]).dt.floor("D").astype(np.int64) / 10**6
-            )
+        checking_eval_set_df = date_converter.convert(checking_eval_set_df)
 
         psi_values_sparse = calculate_sparsity_psi(
             checking_eval_set_df, cat_features, date_column, self.logger, model_task_type

{upgini-1.2.121a3 → upgini-1.2.122a2}/src/upgini/resource_bundle/strings.properties

@@ -155,7 +155,7 @@ target_outliers_warning=We detected {} outliers in your sample.\nExamples of out
     # features validation
 empty_or_contant_features=Columns {} has value with frequency more than 99%, removed from X
 high_cardinality_features=Columns {} has high cardinality (>90% unique values), removed from X
-one_hot_encoded_features=One hot encoded features detected: {}
+one_hot_encoded_features=One hot encoded features detected. Use int encoding for correct results of fit.\n{}
 
     # Dataset validation
 dataset_too_few_rows=X size should be at least {} rows after validation

{upgini-1.2.121a3 → upgini-1.2.122a2}/src/upgini/utils/features_validator.py

@@ -46,7 +46,7 @@ class FeaturesValidator:
 
         if one_hot_encoded_features:
             msg = bundle.get("one_hot_encoded_features").format(one_hot_encoded_features)
-            self.logger.info(msg)
+            warnings.append(msg)
 
         columns_renaming = columns_renaming or {}
 
@@ -100,6 +100,10 @@ class FeaturesValidator:
     @staticmethod
     def is_one_hot_encoded(series: pd.Series) -> bool:
         try:
+            # All rows should be the same type
+            if series.apply(lambda x: type(x)).nunique() != 1:
+                return False
+
             # First, handle string representations of True/False
             series_copy = series.copy()
             if series_copy.dtype == "object" or series_copy.dtype == "string":

{upgini-1.2.121a3 → upgini-1.2.122a2}/src/upgini/utils/psi.py

@@ -7,7 +7,6 @@ from typing import Callable, Dict, Optional
 import more_itertools
 import numpy as np
 import pandas as pd
-from pandas.api.types import is_numeric_dtype
 from pydantic import BaseModel
 
 from upgini.metadata import TARGET, ModelTaskType

upgini-1.2.121a3/src/upgini/__about__.py

@@ -1 +0,0 @@
-__version__ = "1.2.121a3"