pheval 0.4.7__py3-none-any.whl → 0.5.1__py3-none-any.whl

pheval/analyse/binary_classification_stats.py

@@ -1,329 +1,186 @@
-from dataclasses import dataclass, field
-from math import sqrt
-from typing import List, Union
+from dataclasses import dataclass
+from multiprocessing.util import get_logger
 
-from pheval.post_processing.post_processing import (
-    RankedPhEvalDiseaseResult,
-    RankedPhEvalGeneResult,
-    RankedPhEvalVariantResult,
-)
+import polars as pl
 
 
-@dataclass
-class BinaryClassificationStats:
+@dataclass(frozen=True)
+class ConfusionMatrix:
     """
-    A data class representing counts of different categories in binary classification.
+    Define logical conditions for computing a confusion matrix using Polars expressions.
 
     Attributes:
-        true_positives (int): The count of true positive instances - i.e., the number of known entities
-            ranked 1 in the results.
-        true_negatives (int): The count of true negative instances - i.e., the number of non-relevant entities
-            ranked at a position other than 1 in the results.
-        false_positives (int): The count of false positive instances - i.e., the number of non-relevant entities
-            ranked at position 1 in the results.
-        false_negatives (int): The count of false negative instances - i.e., the number of known entities
-            ranked at a position other than 1 in the results.
+        TRUE_POSITIVES (pl.Expr): Condition identifying true positive cases,
+            where `rank == 1` and `true_positive` is `True`.
+        FALSE_POSITIVES (pl.Expr): Condition identifying false positive cases,
+            where `rank == 1` and `true_positive` is `False`.
+        TRUE_NEGATIVES (pl.Expr): Condition identifying true negative cases,
+            where `rank != 1` and `true_positive` is `False`.
+        FALSE_NEGATIVES (pl.Expr): Condition identifying false negative cases,
+            where `rank != 1` and `true_positive` is `True`.
     """
 
-    true_positives: int = 0
-    true_negatives: int = 0
-    false_positives: int = 0
-    false_negatives: int = 0
-    labels: List = field(default_factory=list)
-    scores: List = field(default_factory=list)
-
-    @staticmethod
-    def remove_relevant_ranks(
-        pheval_results: Union[
-            List[RankedPhEvalGeneResult],
-            List[RankedPhEvalVariantResult],
-            List[RankedPhEvalDiseaseResult],
-        ],
-        relevant_ranks: List[int],
-    ) -> List[int]:
-        """
-        Remove the relevant entity ranks from all result ranks
-        Args:
-            pheval_results:
-                (Union[List[RankedPhEvalGeneResult], List[RankedPhEvalVariantResult], List[RankedPhEvalDiseaseResult]]):
-                    The list of all pheval results.
-            relevant_ranks (List[int]): A list of the ranks associated with the known entities.
-
-        Returns:
-            List[int]: A list of the ranks with the relevant entity ranks removed.
-
-        """
-        all_result_ranks = [pheval_result.rank for pheval_result in pheval_results]
-        for rank in relevant_ranks:
-            if rank in all_result_ranks:
-                all_result_ranks.remove(rank)
-                continue
-        return all_result_ranks
-
-    def add_classification_for_known_entities(self, relevant_ranks: List[int]) -> None:
-        """
-        Update binary classification metrics for known entities based on their ranking.
-
-        Args:
-            relevant_ranks (List[int]): A list of the ranks associated with the known entities.
-        """
-        for rank in relevant_ranks:
-            if rank == 1:
-                self.true_positives += 1
-            elif rank != 1:
-                self.false_negatives += 1
-
-    def add_classification_for_other_entities(self, ranks: List[int]) -> None:
-        """
-        Update binary classification metrics for other entities based on their ranking.
-
-        Args:
-            ranks (List[int]): A list of the ranks for all other entities.
-        """
-        for rank in ranks:
-            if rank == 1:
-                self.false_positives += 1
-            elif rank != 1:
-                self.true_negatives += 1
-
-    def add_labels_and_scores(
-        self,
-        pheval_results: Union[
-            List[RankedPhEvalGeneResult],
-            List[RankedPhEvalVariantResult],
-            List[RankedPhEvalDiseaseResult],
-        ],
-        relevant_ranks: List[int],
-    ):
-        """
-        Adds scores and labels from the PhEval results.
-
-        Args:
-            pheval_results (Union[List[RankedPhEvalGeneResult], List[RankedPhEvalVariantResult],
-                                  List[RankedPhEvalDiseaseResult]]):
-                List of all PhEval results
-            relevant_ranks (List[int]): A list of the ranks associated with the known entities.
-        """
-        relevant_ranks_copy = relevant_ranks.copy()
-        for result in pheval_results:
-            self.scores.append(result.score)
-            label = 1 if result.rank in relevant_ranks_copy else 0
-            self.labels.append(label)
-            relevant_ranks_copy.remove(result.rank) if label == 1 else None
-
-    def add_classification(
-        self,
-        pheval_results: Union[
-            List[RankedPhEvalGeneResult],
-            List[RankedPhEvalVariantResult],
-            List[RankedPhEvalDiseaseResult],
-        ],
-        relevant_ranks: List[int],
-    ) -> None:
-        """
-        Update binary classification metrics for known and unknown entities based on their ranks.
-        Args:
-            pheval_results:
-                (Union[List[RankedPhEvalGeneResult], List[RankedPhEvalVariantResult], List[RankedPhEvalDiseaseResult]]):
-                    The list of all pheval results.
-            relevant_ranks (List[int]): A list of the ranks associated with the known entities.
-        """
-        self.add_classification_for_known_entities(relevant_ranks)
-        self.add_classification_for_other_entities(
-            self.remove_relevant_ranks(pheval_results, relevant_ranks)
-        )
-        self.add_labels_and_scores(pheval_results, relevant_ranks)
-
-    def sensitivity(self) -> float:
-        """
-        Calculate sensitivity.
-
-        Sensitivity measures the proportion of actual positive instances correctly identified by the model.
-
-        Returns:
-            float: The sensitivity of the model, calculated as true positives divided by the sum of true positives
-            and false negatives. Returns 0 if both true positives and false negatives are zero.
-        """
-        return (
-            self.true_positives / (self.true_positives + self.false_negatives)
-            if (self.true_positives + self.false_negatives) > 0
-            else 0.0
-        )
-
-    def specificity(self) -> float:
-        """
-        Calculate specificity.
-
-        Specificity measures the proportion of actual negative instances correctly identified by the model.
-
-        Returns:
-            float: The specificity of the model, calculated as true negatives divided by the sum of true negatives
-            and false positives. Returns 0.0 if both true negatives and false positives are zero.
-        """
-        return (
-            self.true_negatives / (self.true_negatives + self.false_positives)
-            if (self.true_negatives + self.false_positives) > 0
-            else 0.0
-        )
-
-    def precision(self) -> float:
-        """
-        Calculate precision.
+    TRUE_POSITIVES = (pl.col("rank") == 1) & (pl.col("true_positive"))
+    FALSE_POSITIVES = (pl.col("rank") == 1) & (~pl.col("true_positive"))
+    TRUE_NEGATIVES = (pl.col("rank") != 1) & (~pl.col("true_positive"))
+    FALSE_NEGATIVES = (pl.col("rank") != 1) & (pl.col("true_positive"))
 
-        Precision measures the proportion of correctly predicted positive instances out of all instances
-        predicted as positive.
-
-        Returns:
-            float: The precision of the model, calculated as true positives divided by the sum of true positives
-            and false positives. Returns 0.0 if both true positives and false positives are zero.
-        """
-        return (
-            self.true_positives / (self.true_positives + self.false_positives)
-            if (self.true_positives + self.false_positives) > 0
-            else 0.0
-        )
 
-    def negative_predictive_value(self) -> float:
-        """
-        Calculate Negative Predictive Value (NPV).
-
-        NPV measures the proportion of correctly predicted negative instances out of all instances predicted negative.
-
-        Returns:
-            float: The Negative Predictive Value of the model, calculated as true negatives divided by the sum of
-            true negatives and false negatives. Returns 0.0 if both true negatives and false negatives are zero.
-        """
-        return (
-            self.true_negatives / (self.true_negatives + self.false_negatives)
-            if (self.true_negatives + self.false_negatives) > 0
-            else 0.0
+@dataclass(frozen=True)
+class BinaryClassificationStats:
+    """Binary classification statistic expressions."""
+
+    SENSITIVITY = (
+        pl.when((pl.col("true_positives") + pl.col("false_negatives")) != 0)
+        .then(pl.col("true_positives") / (pl.col("true_positives") + pl.col("false_negatives")))
+        .otherwise(0.0)
+        .alias("sensitivity")
+    )
+
+    SPECIFICITY = (
+        pl.when((pl.col("true_negatives") + pl.col("false_positives")) != 0)
+        .then(pl.col("true_negatives") / (pl.col("true_negatives") + pl.col("false_positives")))
+        .otherwise(0.0)
+        .alias("specificity")
+    )
+
+    PRECISION = (
+        pl.when((pl.col("true_positives") + pl.col("false_positives")) != 0)
+        .then(pl.col("true_positives") / (pl.col("true_positives") + pl.col("false_positives")))
+        .otherwise(0.0)
+        .alias("precision")
+    )
+
+    NEGATIVE_PREDICTIVE_VALUE = (
+        pl.when((pl.col("true_negatives") + pl.col("false_negatives")) != 0)
+        .then(pl.col("true_negatives") / (pl.col("true_negatives") + pl.col("false_negatives")))
+        .otherwise(0.0)
+        .alias("negative_predictive_value")
+    )
+
+    FALSE_POSITIVE_RATE = (
+        pl.when((pl.col("false_positives") + pl.col("true_negatives")) != 0)
+        .then(pl.col("false_positives") / (pl.col("false_positives") + pl.col("true_negatives")))
+        .otherwise(0.0)
+        .alias("false_positive_rate")
+    )
+
+    FALSE_DISCOVERY_RATE = (
+        pl.when((pl.col("false_positives") + pl.col("true_positives")) != 0)
+        .then(pl.col("false_positives") / (pl.col("false_positives") + pl.col("true_positives")))
+        .otherwise(0.0)
+        .alias("false_discovery_rate")
+    )
+
+    FALSE_NEGATIVE_RATE = (
+        pl.when((pl.col("false_negatives") + pl.col("true_positives")) != 0)
+        .then(pl.col("false_negatives") / (pl.col("false_negatives") + pl.col("true_positives")))
+        .otherwise(0.0)
+        .alias("false_negative_rate")
+    )
+
+    ACCURACY = (
+        pl.when(
+            (
+                pl.col("true_positives")
+                + pl.col("false_positives")
+                + pl.col("true_negatives")
+                + pl.col("false_negatives")
+            )
+            != 0
         )
-
-    def false_positive_rate(self) -> float:
-        """
-        Calculate False Positive Rate (FPR).
-
-        FPR measures the proportion of instances predicted as positive that are actually negative.
-
-        Returns:
-            float: The False Positive Rate of the model, calculated as false positives divided by the sum of
-            false positives and true negatives. Returns 0.0 if both false positives and true negatives are zero.
-        """
-        return (
-            self.false_positives / (self.false_positives + self.true_negatives)
-            if (self.false_positives + self.true_negatives) > 0
-            else 0.0
+        .then(
+            (pl.col("true_positives") + pl.col("true_negatives"))
+            / (
+                pl.col("true_positives")
+                + pl.col("false_positives")
+                + pl.col("true_negatives")
+                + pl.col("false_negatives")
+            )
         )
-
-    def false_discovery_rate(self) -> float:
-        """
-        Calculate False Discovery Rate (FDR).
-
-        FDR measures the proportion of instances predicted as positive that are actually negative.
-
-        Returns:
-            float: The False Discovery Rate of the model, calculated as false positives divided by the sum of
-            false positives and true positives. Returns 0.0 if both false positives and true positives are zero.
-        """
-        return (
-            self.false_positives / (self.false_positives + self.true_positives)
-            if (self.false_positives + self.true_positives) > 0
-            else 0.0
+        .otherwise(0.0)
+        .alias("accuracy")
+    )
+
+    F1_SCORE = (
+        pl.when(
+            2 * (pl.col("true_positives") + pl.col("false_positives") + pl.col("false_negatives"))
+            != 0
         )
-
-    def false_negative_rate(self) -> float:
-        """
-        Calculate False Negative Rate (FNR).
-
-        FNR measures the proportion of instances that are actually positive but predicted as negative.
-
-        Returns:
-            float: The False Negative Rate of the model, calculated as false negatives divided by the sum of
-            false negatives and true positives. Returns 0.0 if both false negatives and true positives are zero.
-        """
-        return (
-            self.false_negatives / (self.false_negatives + self.true_positives)
-            if (self.false_negatives + self.true_positives) > 0
-            else 0.0
+        .then(
+            2
+            * pl.col("true_positives")
+            / (2 * pl.col("true_positives") + pl.col("false_positives") + pl.col("false_negatives"))
         )
+        .otherwise(0.0)
+        .alias("f1_score")
+    )
 
-    def accuracy(self) -> float:
-        """
-        Calculate Accuracy.
-
-        Accuracy measures the proportion of correctly predicted instances out of all instances.
-
-        Returns:
-            float: The Accuracy of the model, calculated as the sum of true positives and true negatives divided by
-            the sum of true positives, false positives, true negatives, and false negatives.
-            Returns 0.0 if the total sum of counts is zero.
-        """
-        return (
-            (self.true_positives + self.true_negatives)
-            / (
-                self.true_positives
-                + self.false_positives
-                + self.true_negatives
-                + self.false_negatives
-            )
-            if (
-                self.true_positives
-                + self.false_negatives
-                + self.true_negatives
-                + self.false_negatives
+    MATTHEWS_CORRELATION_COEFFICIENT = (
+        pl.when(
+            (
+                (pl.col("true_positives") + pl.col("false_positives"))
+                * (pl.col("true_positives") + pl.col("false_negatives"))
+                * (pl.col("true_negatives") + pl.col("false_positives"))
+                * (pl.col("true_negatives") + pl.col("false_negatives"))
             )
             > 0
-            else 0.0
-        )
-
-    def f1_score(self) -> float:
-        """
-        Calculate F1 Score.
-
-        F1 Score is the harmonic mean of precision and recall, providing a balance between false positives
-        and false negatives.
-
-        Returns:
-            float: The F1 Score of the model, calculated as 2 * TP / (2 * TP + FP + FN).
-            Returns 0.0 if the denominator is zero.
-        """
-        return (
-            (2 * self.true_positives)
-            / ((2 * self.true_positives) + self.false_positives + self.false_negatives)
-            if (self.true_positives + self.false_positives + self.false_negatives) > 0
-            else 0.0
         )
-
-    def matthews_correlation_coefficient(self) -> float:
-        """
-        Calculate Matthews Correlation Coefficient (MCC).
-
-        MCC is a measure of the quality of binary classifications, accounting for imbalances in the data.
-
-        Returns:
-            float: The Matthews Correlation Coefficient of the model, calculated as
-            ((TP * TN) - (FP * FN)) / sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN)).
-            Returns 0.0 if the denominator is zero.
-        """
-        return (
+        .then(
             (
-                (self.true_positives * self.true_negatives)
-                - (self.false_positives * self.false_negatives)
+                (pl.col("true_positives") * pl.col("true_negatives"))
+                - (pl.col("false_positives") * pl.col("false_negatives"))
             )
             / (
-                sqrt(
-                    (self.true_positives + self.false_positives)
-                    * (self.true_positives + self.false_negatives)
-                    * (self.true_negatives + self.false_positives)
-                    * (self.true_negatives + self.false_negatives)
-                )
-            )
-            if (
-                self.true_positives
-                + self.false_negatives
-                + self.true_negatives
-                + self.false_negatives
-            )
-            > 0
-            else 0.0
+                (pl.col("true_positives") + pl.col("false_positives"))
+                * (pl.col("true_positives") + pl.col("false_negatives"))
+                * (pl.col("true_negatives") + pl.col("false_positives"))
+                * (pl.col("true_negatives") + pl.col("false_negatives"))
+            ).sqrt()
         )
+        .otherwise(0.0)
+        .alias("matthews_correlation_coefficient")
+    )
+
+
+def compute_confusion_matrix(run_identifier: str, result_scan: pl.LazyFrame) -> pl.LazyFrame:
+    """
+    Computes binary classification statistics.
+
+    Args:
+        run_identifier (str): The identifier for the run.
+        result_scan (pl.LazyFrame): The LazyFrame containing the results for the directory.
+
+    Returns:
+        pl.LazyFrame: The LazyFrame containing the binary classification statistics.
+    """
+    logger = get_logger()
+    logger.info(f"Computing binary classification statistics for {run_identifier}")
+    confusion_matrix = result_scan.select(
+        [
+            pl.lit(run_identifier).alias("run_identifier"),
+            ConfusionMatrix.TRUE_POSITIVES.sum().alias("true_positives").cast(pl.Int64),
+            ConfusionMatrix.FALSE_POSITIVES.sum().alias("false_positives").cast(pl.Int64),
+            ConfusionMatrix.TRUE_NEGATIVES.sum().alias("true_negatives").cast(pl.Int64),
+            ConfusionMatrix.FALSE_NEGATIVES.sum().alias("false_negatives").cast(pl.Int64),
+        ]
+    )
+    return confusion_matrix.select(
+        [
+            pl.col("run_identifier"),
+            pl.col("true_positives"),
+            pl.col("false_positives"),
+            pl.col("true_negatives"),
+            pl.col("false_negatives"),
+            BinaryClassificationStats.SENSITIVITY,
+            BinaryClassificationStats.SPECIFICITY,
+            BinaryClassificationStats.PRECISION,
+            BinaryClassificationStats.NEGATIVE_PREDICTIVE_VALUE,
+            BinaryClassificationStats.FALSE_POSITIVE_RATE,
+            BinaryClassificationStats.FALSE_DISCOVERY_RATE,
+            BinaryClassificationStats.FALSE_NEGATIVE_RATE,
+            BinaryClassificationStats.ACCURACY,
+            BinaryClassificationStats.F1_SCORE,
+            BinaryClassificationStats.MATTHEWS_CORRELATION_COEFFICIENT,
+        ]
+    )