evaluation.py

#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements.  See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License.  You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#

from typing import Generic, List, Optional, Tuple, TypeVar

import sys

from pyspark import since
from pyspark.rdd import RDD
from pyspark.mllib.common import JavaModelWrapper, callMLlibFunc
from pyspark.mllib.linalg import Matrix
from pyspark.sql import SQLContext
from pyspark.sql.types import ArrayType, DoubleType, StructField, StructType

__all__ = [
    "BinaryClassificationMetrics",
    "RegressionMetrics",
    "MulticlassMetrics",
    "RankingMetrics",
]

T = TypeVar("T")


class BinaryClassificationMetrics(JavaModelWrapper):
    """
    Evaluator for binary classification.

    .. versionadded:: 1.4.0

    Parameters
    ----------
    scoreAndLabels : :py:class:`pyspark.RDD`
        an RDD of score, label and optional weight.

    Examples
    --------
    >>> scoreAndLabels = sc.parallelize([
    ...     (0.1, 0.0), (0.1, 1.0), (0.4, 0.0), (0.6, 0.0), (0.6, 1.0), (0.6, 1.0), (0.8, 1.0)], 2)
    >>> metrics = BinaryClassificationMetrics(scoreAndLabels)
    >>> metrics.areaUnderROC
    0.70...
    >>> metrics.areaUnderPR
    0.83...
    >>> metrics.unpersist()
    >>> scoreAndLabelsWithOptWeight = sc.parallelize([
    ...     (0.1, 0.0, 1.0), (0.1, 1.0, 0.4), (0.4, 0.0, 0.2), (0.6, 0.0, 0.6), (0.6, 1.0, 0.9),
    ...     (0.6, 1.0, 0.5), (0.8, 1.0, 0.7)], 2)
    >>> metrics = BinaryClassificationMetrics(scoreAndLabelsWithOptWeight)
    >>> metrics.areaUnderROC
    0.79...
    >>> metrics.areaUnderPR
    0.88...
    """

    def __init__(self, scoreAndLabels: RDD[Tuple[float, float]]):
        sc = scoreAndLabels.ctx
        sql_ctx = SQLContext.getOrCreate(sc)
        numCol = len(scoreAndLabels.first())
        schema = StructType(
            [
                StructField("score", DoubleType(), nullable=False),
                StructField("label", DoubleType(), nullable=False),
            ]
        )
        if numCol == 3:
            schema.add("weight", DoubleType(), False)
        df = sql_ctx.createDataFrame(scoreAndLabels, schema=schema)
        assert sc._jvm is not None
        java_class = sc._jvm.org.apache.spark.mllib.evaluation.BinaryClassificationMetrics
        java_model = java_class(df._jdf)
        super(BinaryClassificationMetrics, self).__init__(java_model)

    @property  # type: ignore[misc]
    @since("1.4.0")
    def areaUnderROC(self) -> float:
        """
        Computes the area under the receiver operating characteristic
        (ROC) curve.
        """
        return self.call("areaUnderROC")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def areaUnderPR(self) -> float:
        """
        Computes the area under the precision-recall curve.
        """
        return self.call("areaUnderPR")

    @since("1.4.0")
    def unpersist(self) -> None:
        """
        Unpersists intermediate RDDs used in the computation.
        """
        self.call("unpersist")


class RegressionMetrics(JavaModelWrapper):
    """
    Evaluator for regression.

    .. versionadded:: 1.4.0

    Parameters
    ----------
    predictionAndObservations : :py:class:`pyspark.RDD`
        an RDD of prediction, observation and optional weight.

    Examples
    --------
    >>> predictionAndObservations = sc.parallelize([
    ...     (2.5, 3.0), (0.0, -0.5), (2.0, 2.0), (8.0, 7.0)])
    >>> metrics = RegressionMetrics(predictionAndObservations)
    >>> metrics.explainedVariance
    8.859...
    >>> metrics.meanAbsoluteError
    0.5...
    >>> metrics.meanSquaredError
    0.37...
    >>> metrics.rootMeanSquaredError
    0.61...
    >>> metrics.r2
    0.94...
    >>> predictionAndObservationsWithOptWeight = sc.parallelize([
    ...     (2.5, 3.0, 0.5), (0.0, -0.5, 1.0), (2.0, 2.0, 0.3), (8.0, 7.0, 0.9)])
    >>> metrics = RegressionMetrics(predictionAndObservationsWithOptWeight)
    >>> metrics.rootMeanSquaredError
    0.68...
    """

    def __init__(self, predictionAndObservations: RDD[Tuple[float, float]]):
        sc = predictionAndObservations.ctx
        sql_ctx = SQLContext.getOrCreate(sc)
        numCol = len(predictionAndObservations.first())
        schema = StructType(
            [
                StructField("prediction", DoubleType(), nullable=False),
                StructField("observation", DoubleType(), nullable=False),
            ]
        )
        if numCol == 3:
            schema.add("weight", DoubleType(), False)
        df = sql_ctx.createDataFrame(predictionAndObservations, schema=schema)
        assert sc._jvm is not None
        java_class = sc._jvm.org.apache.spark.mllib.evaluation.RegressionMetrics
        java_model = java_class(df._jdf)
        super(RegressionMetrics, self).__init__(java_model)

    @property  # type: ignore[misc]
    @since("1.4.0")
    def explainedVariance(self) -> float:
        r"""
        Returns the explained variance regression score.
        explainedVariance = :math:`1 - \frac{variance(y - \hat{y})}{variance(y)}`
        """
        return self.call("explainedVariance")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def meanAbsoluteError(self) -> float:
        """
        Returns the mean absolute error, which is a risk function corresponding to the
        expected value of the absolute error loss or l1-norm loss.
        """
        return self.call("meanAbsoluteError")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def meanSquaredError(self) -> float:
        """
        Returns the mean squared error, which is a risk function corresponding to the
        expected value of the squared error loss or quadratic loss.
        """
        return self.call("meanSquaredError")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def rootMeanSquaredError(self) -> float:
        """
        Returns the root mean squared error, which is defined as the square root of
        the mean squared error.
        """
        return self.call("rootMeanSquaredError")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def r2(self) -> float:
        """
        Returns R^2^, the coefficient of determination.
        """
        return self.call("r2")


class MulticlassMetrics(JavaModelWrapper):
    """
    Evaluator for multiclass classification.

    .. versionadded:: 1.4.0

    Parameters
    ----------
    predictionAndLabels : :py:class:`pyspark.RDD`
        an RDD of prediction, label, optional weight and optional probability.

    Examples
    --------
    >>> predictionAndLabels = sc.parallelize([(0.0, 0.0), (0.0, 1.0), (0.0, 0.0),
    ...     (1.0, 0.0), (1.0, 1.0), (1.0, 1.0), (1.0, 1.0), (2.0, 2.0), (2.0, 0.0)])
    >>> metrics = MulticlassMetrics(predictionAndLabels)
    >>> metrics.confusionMatrix().toArray()
    array([[ 2.,  1.,  1.],
           [ 1.,  3.,  0.],
           [ 0.,  0.,  1.]])
    >>> metrics.falsePositiveRate(0.0)
    0.2...
    >>> metrics.precision(1.0)
    0.75...
    >>> metrics.recall(2.0)
    1.0...
    >>> metrics.fMeasure(0.0, 2.0)
    0.52...
    >>> metrics.accuracy
    0.66...
    >>> metrics.weightedFalsePositiveRate
    0.19...
    >>> metrics.weightedPrecision
    0.68...
    >>> metrics.weightedRecall
    0.66...
    >>> metrics.weightedFMeasure()
    0.66...
    >>> metrics.weightedFMeasure(2.0)
    0.65...
    >>> predAndLabelsWithOptWeight = sc.parallelize([(0.0, 0.0, 1.0), (0.0, 1.0, 1.0),
    ...      (0.0, 0.0, 1.0), (1.0, 0.0, 1.0), (1.0, 1.0, 1.0), (1.0, 1.0, 1.0), (1.0, 1.0, 1.0),
    ...      (2.0, 2.0, 1.0), (2.0, 0.0, 1.0)])
    >>> metrics = MulticlassMetrics(predAndLabelsWithOptWeight)
    >>> metrics.confusionMatrix().toArray()
    array([[ 2.,  1.,  1.],
           [ 1.,  3.,  0.],
           [ 0.,  0.,  1.]])
    >>> metrics.falsePositiveRate(0.0)
    0.2...
    >>> metrics.precision(1.0)
    0.75...
    >>> metrics.recall(2.0)
    1.0...
    >>> metrics.fMeasure(0.0, 2.0)
    0.52...
    >>> metrics.accuracy
    0.66...
    >>> metrics.weightedFalsePositiveRate
    0.19...
    >>> metrics.weightedPrecision
    0.68...
    >>> metrics.weightedRecall
    0.66...
    >>> metrics.weightedFMeasure()
    0.66...
    >>> metrics.weightedFMeasure(2.0)
    0.65...
    >>> predictionAndLabelsWithProbabilities = sc.parallelize([
    ...      (1.0, 1.0, 1.0, [0.1, 0.8, 0.1]), (0.0, 2.0, 1.0, [0.9, 0.05, 0.05]),
    ...      (0.0, 0.0, 1.0, [0.8, 0.2, 0.0]), (1.0, 1.0, 1.0, [0.3, 0.65, 0.05])])
    >>> metrics = MulticlassMetrics(predictionAndLabelsWithProbabilities)
    >>> metrics.logLoss()
    0.9682...
    """

    def __init__(self, predictionAndLabels: RDD[Tuple[float, float]]):
        sc = predictionAndLabels.ctx
        sql_ctx = SQLContext.getOrCreate(sc)
        numCol = len(predictionAndLabels.first())
        schema = StructType(
            [
                StructField("prediction", DoubleType(), nullable=False),
                StructField("label", DoubleType(), nullable=False),
            ]
        )
        if numCol >= 3:
            schema.add("weight", DoubleType(), False)
        if numCol == 4:
            schema.add("probability", ArrayType(DoubleType(), False), False)
        df = sql_ctx.createDataFrame(predictionAndLabels, schema)
        assert sc._jvm is not None
        java_class = sc._jvm.org.apache.spark.mllib.evaluation.MulticlassMetrics
        java_model = java_class(df._jdf)
        super(MulticlassMetrics, self).__init__(java_model)

    @since("1.4.0")
    def confusionMatrix(self) -> Matrix:
        """
        Returns confusion matrix: predicted classes are in columns,
        they are ordered by class label ascending, as in "labels".
        """
        return self.call("confusionMatrix")

    @since("1.4.0")
    def truePositiveRate(self, label: float) -> float:
        """
        Returns true positive rate for a given label (category).
        """
        return self.call("truePositiveRate", label)

    @since("1.4.0")
    def falsePositiveRate(self, label: float) -> float:
        """
        Returns false positive rate for a given label (category).
        """
        return self.call("falsePositiveRate", label)

    @since("1.4.0")
    def precision(self, label: float) -> float:
        """
        Returns precision.
        """
        return self.call("precision", float(label))

    @since("1.4.0")
    def recall(self, label: float) -> float:
        """
        Returns recall.
        """
        return self.call("recall", float(label))

    @since("1.4.0")
    def fMeasure(self, label: float, beta: Optional[float] = None) -> float:
        """
        Returns f-measure.
        """
        if beta is None:
            return self.call("fMeasure", label)
        else:
            return self.call("fMeasure", label, beta)

    @property  # type: ignore[misc]
    @since("2.0.0")
    def accuracy(self) -> float:
        """
        Returns accuracy (equals to the total number of correctly classified instances
        out of the total number of instances).
        """
        return self.call("accuracy")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def weightedTruePositiveRate(self) -> float:
        """
        Returns weighted true positive rate.
        (equals to precision, recall and f-measure)
        """
        return self.call("weightedTruePositiveRate")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def weightedFalsePositiveRate(self) -> float:
        """
        Returns weighted false positive rate.
        """
        return self.call("weightedFalsePositiveRate")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def weightedRecall(self) -> float:
        """
        Returns weighted averaged recall.
        (equals to precision, recall and f-measure)
        """
        return self.call("weightedRecall")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def weightedPrecision(self) -> float:
        """
        Returns weighted averaged precision.
        """
        return self.call("weightedPrecision")

    @since("1.4.0")
    def weightedFMeasure(self, beta: Optional[float] = None) -> float:
        """
        Returns weighted averaged f-measure.
        """
        if beta is None:
            return self.call("weightedFMeasure")
        else:
            return self.call("weightedFMeasure", beta)

    @since("3.0.0")
    def logLoss(self, eps: float = 1e-15) -> float:
        """
        Returns weighted logLoss.
        """
        return self.call("logLoss", eps)


class RankingMetrics(JavaModelWrapper, Generic[T]):
    """
    Evaluator for ranking algorithms.

    .. versionadded:: 1.4.0

    Parameters
    ----------
    predictionAndLabels : :py:class:`pyspark.RDD`
        an RDD of (predicted ranking, ground truth set) pairs.

    Examples
    --------
    >>> predictionAndLabels = sc.parallelize([
    ...     ([1, 6, 2, 7, 8, 3, 9, 10, 4, 5], [1, 2, 3, 4, 5]),
    ...     ([4, 1, 5, 6, 2, 7, 3, 8, 9, 10], [1, 2, 3]),
    ...     ([1, 2, 3, 4, 5], [])])
    >>> metrics = RankingMetrics(predictionAndLabels)
    >>> metrics.precisionAt(1)
    0.33...
    >>> metrics.precisionAt(5)
    0.26...
    >>> metrics.precisionAt(15)
    0.17...
    >>> metrics.meanAveragePrecision
    0.35...
    >>> metrics.meanAveragePrecisionAt(1)
    0.3333333333333333...
    >>> metrics.meanAveragePrecisionAt(2)
    0.25...
    >>> metrics.ndcgAt(3)
    0.33...
    >>> metrics.ndcgAt(10)
    0.48...
    >>> metrics.recallAt(1)
    0.06...
    >>> metrics.recallAt(5)
    0.35...
    >>> metrics.recallAt(15)
    0.66...
    """

    def __init__(self, predictionAndLabels: RDD[Tuple[List[T], List[T]]]):
        sc = predictionAndLabels.ctx
        sql_ctx = SQLContext.getOrCreate(sc)
        df = sql_ctx.createDataFrame(
            predictionAndLabels, schema=sql_ctx._inferSchema(predictionAndLabels)
        )
        java_model = callMLlibFunc("newRankingMetrics", df._jdf)
        super(RankingMetrics, self).__init__(java_model)

    @since("1.4.0")
    def precisionAt(self, k: int) -> float:
        """
        Compute the average precision of all the queries, truncated at ranking position k.

        If for a query, the ranking algorithm returns n (n < k) results, the precision value
        will be computed as #(relevant items retrieved) / k. This formula also applies when
        the size of the ground truth set is less than k.

        If a query has an empty ground truth set, zero will be used as precision together
        with a log warning.
        """
        return self.call("precisionAt", int(k))

    @property  # type: ignore[misc]
    @since("1.4.0")
    def meanAveragePrecision(self) -> float:
        """
        Returns the mean average precision (MAP) of all the queries.
        If a query has an empty ground truth set, the average precision will be zero and
        a log warning is generated.
        """
        return self.call("meanAveragePrecision")

    @since("3.0.0")
    def meanAveragePrecisionAt(self, k: int) -> float:
        """
        Returns the mean average precision (MAP) at first k ranking of all the queries.
        If a query has an empty ground truth set, the average precision will be zero and
        a log warning is generated.
        """
        return self.call("meanAveragePrecisionAt", int(k))

    @since("1.4.0")
    def ndcgAt(self, k: int) -> float:
        """
        Compute the average NDCG value of all the queries, truncated at ranking position k.
        The discounted cumulative gain at position k is computed as:
        sum,,i=1,,^k^ (2^{relevance of ''i''th item}^ - 1) / log(i + 1),
        and the NDCG is obtained by dividing the DCG value on the ground truth set.
        In the current implementation, the relevance value is binary.
        If a query has an empty ground truth set, zero will be used as NDCG together with
        a log warning.
        """
        return self.call("ndcgAt", int(k))

    @since("3.0.0")
    def recallAt(self, k: int) -> float:
        """
        Compute the average recall of all the queries, truncated at ranking position k.

        If for a query, the ranking algorithm returns n results, the recall value
        will be computed as #(relevant items retrieved) / #(ground truth set).
        This formula also applies when the size of the ground truth set is less than k.

        If a query has an empty ground truth set, zero will be used as recall together
        with a log warning.
        """
        return self.call("recallAt", int(k))


class MultilabelMetrics(JavaModelWrapper):
    """
    Evaluator for multilabel classification.

    .. versionadded:: 1.4.0

    Parameters
    ----------
    predictionAndLabels : :py:class:`pyspark.RDD`
        an RDD of (predictions, labels) pairs,
        both are non-null Arrays, each with unique elements.

    Examples
    --------
    >>> predictionAndLabels = sc.parallelize([([0.0, 1.0], [0.0, 2.0]), ([0.0, 2.0], [0.0, 1.0]),
    ...     ([], [0.0]), ([2.0], [2.0]), ([2.0, 0.0], [2.0, 0.0]),
    ...     ([0.0, 1.0, 2.0], [0.0, 1.0]), ([1.0], [1.0, 2.0])])
    >>> metrics = MultilabelMetrics(predictionAndLabels)
    >>> metrics.precision(0.0)
    1.0
    >>> metrics.recall(1.0)
    0.66...
    >>> metrics.f1Measure(2.0)
    0.5
    >>> metrics.precision()
    0.66...
    >>> metrics.recall()
    0.64...
    >>> metrics.f1Measure()
    0.63...
    >>> metrics.microPrecision
    0.72...
    >>> metrics.microRecall
    0.66...
    >>> metrics.microF1Measure
    0.69...
    >>> metrics.hammingLoss
    0.33...
    >>> metrics.subsetAccuracy
    0.28...
    >>> metrics.accuracy
    0.54...
    """

    def __init__(self, predictionAndLabels: RDD[Tuple[List[float], List[float]]]):
        sc = predictionAndLabels.ctx
        sql_ctx = SQLContext.getOrCreate(sc)
        df = sql_ctx.createDataFrame(
            predictionAndLabels, schema=sql_ctx._inferSchema(predictionAndLabels)
        )
        assert sc._jvm is not None
        java_class = sc._jvm.org.apache.spark.mllib.evaluation.MultilabelMetrics
        java_model = java_class(df._jdf)
        super(MultilabelMetrics, self).__init__(java_model)

    @since("1.4.0")
    def precision(self, label: Optional[float] = None) -> float:
        """
        Returns precision or precision for a given label (category) if specified.
        """
        if label is None:
            return self.call("precision")
        else:
            return self.call("precision", float(label))

    @since("1.4.0")
    def recall(self, label: Optional[float] = None) -> float:
        """
        Returns recall or recall for a given label (category) if specified.
        """
        if label is None:
            return self.call("recall")
        else:
            return self.call("recall", float(label))

    @since("1.4.0")
    def f1Measure(self, label: Optional[float] = None) -> float:
        """
        Returns f1Measure or f1Measure for a given label (category) if specified.
        """
        if label is None:
            return self.call("f1Measure")
        else:
            return self.call("f1Measure", float(label))

    @property  # type: ignore[misc]
    @since("1.4.0")
    def microPrecision(self) -> float:
        """
        Returns micro-averaged label-based precision.
        (equals to micro-averaged document-based precision)
        """
        return self.call("microPrecision")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def microRecall(self) -> float:
        """
        Returns micro-averaged label-based recall.
        (equals to micro-averaged document-based recall)
        """
        return self.call("microRecall")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def microF1Measure(self) -> float:
        """
        Returns micro-averaged label-based f1-measure.
        (equals to micro-averaged document-based f1-measure)
        """
        return self.call("microF1Measure")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def hammingLoss(self) -> float:
        """
        Returns Hamming-loss.
        """
        return self.call("hammingLoss")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def subsetAccuracy(self) -> float:
        """
        Returns subset accuracy.
        (for equal sets of labels)
        """
        return self.call("subsetAccuracy")

    @property  # type: ignore[misc]
    @since("1.4.0")
    def accuracy(self) -> float:
        """
        Returns accuracy.
        """
        return self.call("accuracy")


def _test() -> None:
    import doctest
    import numpy
    from pyspark.sql import SparkSession
    import pyspark.mllib.evaluation

    try:
        # Numpy 1.14+ changed it's string format.
        numpy.set_printoptions(legacy="1.13")
    except TypeError:
        pass
    globs = pyspark.mllib.evaluation.__dict__.copy()
    spark = SparkSession.builder.master("local[4]").appName("mllib.evaluation tests").getOrCreate()
    globs["sc"] = spark.sparkContext
    (failure_count, test_count) = doctest.testmod(globs=globs, optionflags=doctest.ELLIPSIS)
    spark.stop()
    if failure_count:
        sys.exit(-1)


if __name__ == "__main__":
    _test()