random-cut-forest-by-aws/python_rcf_wrapper/trcf_model.py at main · aws/random-cut-forest-by-aws · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
# Java imports
from typing import List, Optional, Tuple, Any

import numpy as np
import logging
from com.amazon.randomcutforest.parkservices import ThresholdedRandomCutForest
from com.amazon.randomcutforest.config import Precision
from com.amazon.randomcutforest.parkservices import AnomalyDescriptor
from com.amazon.randomcutforest.config import TransformMethod
import jpype

class TRandomCutForestModel:
    """
    Random Cut Forest Python Binding around the AWS Random Cut Forest Official Java version:
    https://github.com/aws/random-cut-forest-by-aws
    """

    def __init__(self, rcf_dimensions, shingle_size, num_trees: int = 30, output_after: int=256, anomaly_rate=0.005,
                 z_factor=2.5, score_differencing=0.5, ignore_delta_threshold=0, sample_size=256):
        self.forest = (ThresholdedRandomCutForest
        .builder()
        .dimensions(rcf_dimensions)
        .sampleSize(sample_size)
        .numberOfTrees(num_trees)
        .timeDecay(0.0001)
        .initialAcceptFraction(output_after*1.0/sample_size)
        .parallelExecutionEnabled(True)
        .compact(True)
        .precision(Precision.FLOAT_32)
        .boundingBoxCacheFraction(1)
        .shingleSize(shingle_size)
        .anomalyRate(anomaly_rate)
        .outputAfter(output_after)
        .internalShinglingEnabled(True)
        .transformMethod(TransformMethod.NORMALIZE)
        .alertOnce(True)
        .autoAdjust(True)
        .build())
        self.forest.setZfactor(z_factor)

    def process(self, point: List[float]) -> AnomalyDescriptor:
        """
        a single call that prepreprocesses data, compute score/grade and updates
        state.

        Parameters
        ----------
        point: List[float]
            A data point with shingle size

        Returns
        -------
        AnomalyDescriptor
             Encapsulate detailed information about anomalies detected by RCF model. This class stores various attributes
             related to an anomaly, such as confidence levels, attribution scores, and expected values.

        """
        return self.forest.process(point, 0)