Change metric.input_name -> metric.input_names (plural)

langkit/core/metric.py

@@ -84,7 +84,7 @@ class MultiMetricResult:
 @dataclass(frozen=True)
 class SingleMetric:
     name: str  # Basically the output name
-    input_name: List[str]
+    input_names: List[str]
     evaluate: Callable[[pd.DataFrame], SingleMetricResult]
     init: Optional[Callable[[], None]] = None
     cache_assets: Optional[Callable[[], None]] = None
@@ -95,7 +95,7 @@ class MultiMetric:
     # Splitting the metric into single/multi can be a bit verbose, but it lets us know all of the metric names
     # that are going to be generated upfront without having to evaluate all of the metrics to find out.
     names: List[str]
-    input_name: List[str]
+    input_names: List[str]
     evaluate: Callable[[pd.DataFrame], MultiMetricResult]
     init: Optional[Callable[[], None]] = None
     cache_assets: Optional[Callable[[], None]] = None

langkit/core/workflow.py

@@ -205,14 +205,14 @@ def run(self, data: Union[pd.DataFrame, Row, Dict[str, str]], options: Optional[
 
         if options and options.metric_filter and options.metric_filter.by_required_inputs:
             by_required_inputs_set = frozenset([frozenset(x) for x in options.metric_filter.by_required_inputs])
-            metrics_to_run = [metric for metric in self.metrics_config.metrics if frozenset(metric.input_name) in by_required_inputs_set]
+            metrics_to_run = [metric for metric in self.metrics_config.metrics if frozenset(metric.input_names) in by_required_inputs_set]
         else:
             metrics_to_run = self.metrics_config.metrics
 
         for metric in metrics_to_run:
             # check that the dataframe has the metric.input_name present, or else skip
-            if not all([input_name in df.columns for input_name in metric.input_name]):
-                logger.debug(f"Skipping metric {metric} because {metric.input_name} is not present in the input dataframe")
+            if not all([input_name in df.columns for input_name in metric.input_names]):
+                logger.debug(f"Skipping metric {metric} because {metric.input_names} is not present in the input dataframe")
                 continue
 
             metric_start = time.perf_counter()

langkit/metrics/injections.py

@@ -87,7 +87,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
         return SingleMetricResult(metrics=metrics)
 
     return SingleMetric(
-        name=f"{column_name}.similarity.injection", input_name=[column_name], evaluate=udf, cache_assets=cache_assets, init=init
+        name=f"{column_name}.similarity.injection", input_names=[column_name], evaluate=udf, cache_assets=cache_assets, init=init
     )
 
 

langkit/metrics/input_output_similarity.py

@@ -24,7 +24,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
 
     return SingleMetric(
         name=f"{output_column_name}.similarity.{input_column_name}",
-        input_name=[input_column_name, output_column_name],
+        input_names=[input_column_name, output_column_name],
         evaluate=udf,
         init=init,
     )

langkit/metrics/pii.py

@@ -94,7 +94,7 @@ def process_row(row: pd.DataFrame) -> Dict[str, List[Optional[str]]]:
         return MultiMetricResult(metrics=all_metrics)
 
     metric_names = list(entity_types.values()) + [redacted_metric_name]
-    return MultiMetric(names=metric_names, input_name=[input_name], evaluate=udf, cache_assets=cache_assets, init=init)
+    return MultiMetric(names=metric_names, input_names=[input_name], evaluate=udf, cache_assets=cache_assets, init=init)
 
 
 prompt_presidio_pii_metric = partial(pii_presidio_metric, "prompt")

langkit/metrics/regexes/regexes.py

@@ -61,7 +61,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
 
     return SingleMetric(
         name=f"{column_name}.regex.has_patterns",
-        input_name=[column_name],
+        input_names=[column_name],
         evaluate=udf,
     )
 
@@ -126,7 +126,7 @@ def udf(text: Union[pd.DataFrame, Dict[str, List[Any]]]) -> SingleMetricResult:
 
     return SingleMetric(
         name=f"{column_name}.regex.{__sanitize_name_for_metric(pattern_name)}",
-        input_name=[column_name],
+        input_names=[column_name],
         evaluate=udf,
     )
 
@@ -239,7 +239,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
 
     return SingleMetric(
         name=f"{column_name}.substitutions.{__sanitize_name_for_metric(pattern_name)}",
-        input_name=[column_name],
+        input_names=[column_name],
         evaluate=udf,
     )
 

langkit/metrics/sentiment_polarity.py

@@ -30,7 +30,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
 
     return SingleMetric(
         name=f"{column_name}.sentiment.sentiment_score",
-        input_name=[column_name],
+        input_names=[column_name],
         evaluate=udf,
         init=init,
         cache_assets=cache_assets,

langkit/metrics/text_statistics.py

@@ -15,7 +15,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
 
     return SingleMetric(
         name=f"{column_name}.stats.{stat}",
-        input_name=[column_name],
+        input_names=[column_name],
         evaluate=udf,
     )
 

langkit/metrics/themes/themes.py

@@ -81,7 +81,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
 
     return SingleMetric(
         name=f"{column_name}.similarity.{themes_group}",
-        input_name=[column_name],
+        input_names=[column_name],
         evaluate=udf,
         cache_assets=cache_assets,
     )

langkit/metrics/token.py

@@ -25,7 +25,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
 
     return SingleMetric(
         name=f"{column_name}.stats.token_count",
-        input_name=[column_name],
+        input_names=[column_name],
         evaluate=udf,
         init=init,
         cache_assets=cache_assets,

langkit/metrics/topic.py

@@ -70,7 +70,7 @@ def cache_assets():
         __classifier.value
 
     metric_names = [f"{input_name}.topics.{_sanitize_metric_name(topic)}" for topic in topics]
-    return MultiMetric(names=metric_names, input_name=[input_name], evaluate=udf, cache_assets=cache_assets)
+    return MultiMetric(names=metric_names, input_names=[input_name], evaluate=udf, cache_assets=cache_assets)
 
 
 prompt_topic_module = partial(topic_metric, "prompt", __default_topics, _hypothesis_template)

langkit/metrics/toxicity.py

@@ -51,7 +51,7 @@ def udf(text: pd.DataFrame) -> SingleMetricResult:
         metrics = __toxicity(_pipeline, max_length, col)
         return SingleMetricResult(metrics=metrics)
 
-    return SingleMetric(name=f"{column_name}.toxicity.toxicity_score", input_name=[column_name], evaluate=udf, init=init)
+    return SingleMetric(name=f"{column_name}.toxicity.toxicity_score", input_names=[column_name], evaluate=udf, init=init)
 
 
 prompt_toxicity_metric = partial(toxicity_metric, "prompt")

langkit/metrics/whylogs_compat.py

@@ -63,8 +63,8 @@ def udf(text: Union[pd.DataFrame, Dict[str, List[Any]]]) -> Any:
         types = {"prompt": str, "response": str}
         column_names = ["prompt", "response"]
     else:
-        types = {metric.input_name: str}
-        column_names = [metric.input_name]
+        types = {metric.input_names: str}
+        column_names = [metric.input_names]
 
     schema = UdfSchemaArgs(
         types=types,
@@ -89,7 +89,7 @@ def udf(text: Union[pd.DataFrame, Dict[str, List[Any]]]) -> Any:
     return UdfSchemaArgs(
         resolvers=[],
         types={k: str for k in metric.names},
-        udf_specs=[UdfSpec(column_names=[metric.input_name], udf=udf, prefix="")],
+        udf_specs=[UdfSpec(column_names=[metric.input_names], udf=udf, prefix="")],
     )
 
 

tests/langkit/metrics/test_text_statistics.py

@@ -414,7 +414,7 @@ def udf(text: pd.DataFrame) -> MultiMetricResult:
 
         return MultiMetric(
             names=[f"{column_name}.custom_textstat1", f"{column_name}.custom_textstat2"],
-            input_name=column_name,
+            input_names=column_name,
             evaluate=udf,
         )