Merge pull request #754 from mindsdb/dielectron_problems

Dielectron problems

lightwood/api/types.py

@@ -355,7 +355,7 @@ def from_dict(obj: Dict):
         target_weights = obj.get('target_weights', None)
         positive_domain = obj.get('positive_domain', False)
         timeseries_settings = TimeseriesSettings.from_dict(obj.get('timeseries_settings', {}))
-        anomaly_detection = obj.get('anomaly_detection', True)
+        anomaly_detection = obj.get('anomaly_detection', False)
         ignore_features = obj.get('ignore_features', [])
         fit_on_all = obj.get('fit_on_all', True)
         strict_mode = obj.get('strict_mode', True)

lightwood/data/cleaner.py

@@ -289,7 +289,7 @@ def _remove_columns(data: pd.DataFrame, identifiers: Dict[str, object], target:
     if mode == "predict":
         if (
             target in data.columns
-            and not timeseries_settings.use_previous_target
+            and (not timeseries_settings.is_timeseries or not timeseries_settings.use_previous_target)
             and not anomaly_detection
         ):
             data = data.drop(columns=[target])

lightwood/data/encoded_ds.py

@@ -74,7 +74,8 @@ def __getitem__(self, idx: int) -> Tuple[torch.Tensor, torch.Tensor]:
 
                 encoded_tensor = self.encoders[col].encode(data, **kwargs)[0]
                 if torch.isnan(encoded_tensor).any() or torch.isinf(encoded_tensor).any():
-                    raise Exception(f'Encoded tensor: {encoded_tensor} contains nan or inf values')
+                    raise Exception(f'Encoded tensor: {encoded_tensor} contains nan or inf values, this tensor is \
+                                      the encoding of column {col} using {self.encoders[col].__class__}')
                 if col != self.target:
                     X = torch.cat([X, encoded_tensor])
                 else:

lightwood/data/timeseries_transform.py

@@ -43,6 +43,10 @@ def transform_timeseries(
     if tss.use_previous_target and target not in data.columns:
         raise Exception(f"Cannot transform. Missing historical values for target column {target} (`use_previous_target` is set to True).")  # noqa
 
+    for hcol in tss.historical_columns:
+        if hcol not in data.columns or data[hcol].isna().any():
+            raise Exception(f"Cannot transform. Missing values in historical column {hcol}.")
+
     if '__mdb_make_predictions' in original_df.columns:
         index = original_df[original_df['__mdb_make_predictions'].map(
             {'True': True, 'False': False, True: True, False: False}).isin([True])]

lightwood/helpers/accuracy.py

@@ -0,0 +1,14 @@
+from sklearn.metrics import r2_score as sk_r2_score
+
+
+def r2_score(y_true, y_pred) -> float:
+    """ Wrapper for sklearn R2 score, lower capped between 0 and 1"""
+    acc = sk_r2_score(y_true, y_pred)
+    # Cap at 0
+    if acc < 0:
+        acc = 0
+    # Guard against overflow (> 1 means overflow of negative score)
+    if acc > 1:
+        acc = 0
+
+    return acc

lightwood/helpers/general.py

@@ -65,7 +65,11 @@ def evaluate_accuracy(data: pd.DataFrame,
                                                                         ts_analysis=ts_analysis)
         else:
             true_values = data[target].tolist()
-            accuracy_function = getattr(importlib.import_module('sklearn.metrics'), accuracy_function_str)
+            if hasattr(importlib.import_module('lightwood.helpers.accuracy'), accuracy_function_str):
+                accuracy_function = getattr(importlib.import_module('lightwood.helpers.accuracy'),
+                                            accuracy_function_str)
+            else:
+                accuracy_function = getattr(importlib.import_module('sklearn.metrics'), accuracy_function_str)
             score_dict[accuracy_function_str] = accuracy_function(list(true_values), list(predictions))
 
     return score_dict

tests/integration/advanced/test_timeseries.py

@@ -184,6 +184,7 @@ def test_3_time_series_sktime_mixer(self):
         df = pd.DataFrame(columns=['Time', target])
         df['Time'] = t
         df[target] = ts
+        df[f'{target}_2x'] = 2 * ts
 
         train = df[:int(len(df) * 0.8)]
         test = df[int(len(df) * 0.8):]
@@ -193,7 +194,8 @@ def test_3_time_series_sktime_mixer(self):
                                             'timeseries_settings': {
                                                 'order_by': ['Time'],
                                                 'window': 5,
-                                                'nr_predictions': 20
+                                                'nr_predictions': 20,
+                                                'historical_columns': [f'{target}_2x']
                                             }})
 
         json_ai = json_ai_from_problem(df, problem_definition=pdef)
@@ -209,5 +211,11 @@ def test_3_time_series_sktime_mixer(self):
 
         predictor.learn(train)
         ps = predictor.predict(test)
-
         assert r2_score(ps['truth'].values, ps['prediction'].iloc[0]) >= 0.95
+
+        # test historical columns asserts
+        test[f'{target}_2x'].iloc[0] = np.nan
+        self.assertRaises(Exception, predictor.predict, test)
+
+        test.pop(f'{target}_2x')
+        self.assertRaises(Exception, predictor.predict, test)