bigml/resource/
evaluation.rs

1//! An evaluation of how well a model (or ensemble) predicts the data.
2
3use serde::de::DeserializeOwned;
4use serde::{Deserialize, Serialize};
5use std::fmt;
6
7use super::id::*;
8use super::status::*;
9use super::{Resource, ResourceCommon};
10
11/// An evaluation of how well a model (or ensemble) predicts the data.
12///
13/// TODO: Still lots of missing fields.
14#[derive(Clone, Debug, Deserialize, Resource, Serialize)]
15#[serde(bound(deserialize = ""))]
16#[api_name = "evaluation"]
17#[non_exhaustive]
18pub struct Evaluation<R: Result> {
19    /// Common resource information. These fields will be serialized at the
20    /// top-level of this structure by `serde`.
21    #[serde(flatten)]
22    pub common: ResourceCommon,
23
24    /// The ID of this resource.
25    pub resource: Id<Evaluation<R>>,
26
27    /// The status of this resource.
28    pub status: GenericStatus,
29
30    /// The result of this evaluation.
31    pub result: R,
32}
33
34/// The result of an evaluation.
35///
36/// TODO: I'm not sure we want to shadow `Result`.  But this name will
37/// basically always be qualified, so maybe it's OK.
38pub trait Result:
39    fmt::Debug + DeserializeOwned + Send + Serialize + Sized + Sync + 'static
40{
41}
42
43/// The result of evaluating a classifier.
44#[derive(Debug, Deserialize, Serialize)]
45#[non_exhaustive]
46pub struct ClassificationResult {
47    /// The names of our classifications.
48    pub class_names: Vec<String>,
49
50    /// According to BigML, "Measures the performance of the classifier
51    /// that predicts the mode class for all the instances in the dataset."
52    pub mode: DetailedClassificationResult,
53
54    /// The performance of this model.
55    pub model: DetailedClassificationResult,
56
57    /// According to BigML, "Measures the performance of the classifier
58    /// that predicts a random class for all the instances in the dataset."
59    pub random: DetailedClassificationResult,
60}
61
62impl Result for ClassificationResult {}
63
64/// The detailed result of an evaluation using specific criteria.
65#[derive(Debug, Deserialize, Serialize)]
66#[non_exhaustive]
67pub struct DetailedClassificationResult {
68    /// The portion of instances we classified correctly.
69    pub accuracy: f64,
70    /// The average f-measure over all classes.
71    pub average_f_measure: f64,
72    /// The average phi over all classes.
73    pub average_phi: f64,
74    /// The average precision over all classes.
75    pub average_precision: f64,
76    /// The average recall over all classes.
77    pub average_recall: f64,
78    /// A list of rows of the confusion matrix for this model.
79    pub confusion_matrix: Vec<Vec<f64>>,
80    /// Statistics for each of the individidual classes.
81    pub per_class_statistics: Vec<ClassificationPerClassStatistics>,
82}
83
84/// The detailed result of an evaluation using specific criteria.
85#[derive(Debug, Deserialize, Serialize)]
86#[non_exhaustive]
87pub struct ClassificationPerClassStatistics {
88    /// The portion of instances in this class that were correctly
89    /// classified.
90    pub accuracy: f64,
91    /// The the of this class.
92    pub class_name: String,
93    /// The harmonic mean of precision and recall.
94    pub f_measure: f64,
95    /// See
96    /// [Wikipedia](http://en.wikipedia.org/wiki/Matthews_correlation_coefficient).
97    pub phi_coefficient: f64,
98    /// The fraction of positives that were true positives. (TP / (TP + FP))
99    pub precision: f64,
100    /// The number of true positives over the number of actual positives in
101    /// the dataset. (TP / (TP + FN))
102    pub recall: f64,
103}
104
105// TODO: RegressionResult.
bigml/resource/evaluation.rs

bigml/resource/
evaluation.rs
