fsrs-vs-sm17/utils.py at main · open-spaced-repetition/fsrs-vs-sm17 · 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
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
import matplotlib.pyplot as plt
import numpy as np
from sklearn.metrics import root_mean_squared_error
import pandas as pd


def get_bin(x, bins=10):
    # Simple equal-width binning with 0.1 intervals from 0 to 1
    # This matches the binning method shown in the graph
    return np.round(x * bins) / bins


def cross_comparison(revlogs, algoA, algoB):
    cross_comparison_record = revlogs[[f"R ({algoA})", f"R ({algoB})", "y"]].copy()

    for algo in (algoA, algoB):
        cross_comparison_record[f"{algo}_B-W"] = (
            cross_comparison_record[f"R ({algo})"] - cross_comparison_record["y"]
        )
        cross_comparison_record[f"{algo}_bin"] = cross_comparison_record[
            f"R ({algo})"
        ].map(get_bin)

    fig = plt.figure(figsize=(6, 6))
    ax = fig.gca()
    ax.axhline(y=0.0, color="black", linestyle="-")
    result = {}

    for referee, player in [(algoA, algoB), (algoB, algoA)]:
        cross_comparison_group = cross_comparison_record.groupby(
            by=f"{referee}_bin"
        ).agg(
            {
                "y": ["mean"],
                f"{player}_B-W": ["mean"],
                f"R ({player})": ["mean", "count"],
            }
        )
        universal_metric = root_mean_squared_error(
            cross_comparison_group["y", "mean"],
            cross_comparison_group[f"R ({player})", "mean"],
            sample_weight=cross_comparison_group[f"R ({player})", "count"],
        )
        result[f"{player}_evaluated_by_{referee}"] = round(universal_metric, 4)
        cross_comparison_group[f"R ({player})", "percent"] = (
            cross_comparison_group[f"R ({player})", "count"]
            / cross_comparison_group[f"R ({player})", "count"].sum()
        )
        ax.scatter(
            cross_comparison_group.index,
            cross_comparison_group[f"{player}_B-W", "mean"],
            s=cross_comparison_group[f"R ({player})", "percent"] * 1024,
            alpha=0.5,
        )
        ax.plot(
            cross_comparison_group[f"{player}_B-W", "mean"],
            label=f"{player} by {referee}, UM={universal_metric:.4f}",
        )

    ax.legend(loc="lower center")
    ax.grid(linestyle="--")
    ax.set_title(f"{algoA} vs {algoB}")
    ax.set_xlabel("Predicted R")
    ax.set_ylabel("B-W Metric")
    ax.set_xlim(0, 1)
    ax.set_xticks(np.arange(0, 1.1, 0.1))
    plt.tight_layout()
    plt.show()

    return result


def cross_comparison_plus(revlogs, algoA, algoB):
    cross_comparison_record = revlogs[[f"R ({algoA})", f"R ({algoB})", "y"]].copy()

    for algo in (algoA, algoB):
        cross_comparison_record[f"{algo}_B-W"] = (
            cross_comparison_record[f"R ({algo})"] - cross_comparison_record["y"]
        )
        cross_comparison_record["R_diff"] = (
            cross_comparison_record[f"R ({algoA})"]
            - cross_comparison_record[f"R ({algoB})"]
        )
        cross_comparison_record[f"{algo}_bin"] = cross_comparison_record[f"R_diff"].map(
            lambda x: get_bin(x, bins=20)
        )

    fig = plt.figure(figsize=(12, 6))
    ax = fig.gca()
    ax.axhline(y=0.0, color="black", linestyle="-")
    result = {}

    for referee, player in [(algoA, algoB), (algoB, algoA)]:
        cross_comparison_group = cross_comparison_record.groupby(
            by=f"{referee}_bin"
        ).agg(
            {
                "R_diff": ["mean"],
                "y": ["mean"],
                f"{player}_B-W": ["mean"],
                f"R ({player})": ["mean", "count"],
            }
        )
        universal_metric = root_mean_squared_error(
            cross_comparison_group["y", "mean"],
            cross_comparison_group[f"R ({player})", "mean"],
            sample_weight=cross_comparison_group[f"R ({player})", "count"],
        )
        result[f"{player}_evaluated_by_{referee}"] = round(universal_metric, 4)
        cross_comparison_group[f"R ({player})", "percent"] = (
            cross_comparison_group[f"R ({player})", "count"]
            / cross_comparison_group[f"R ({player})", "count"].sum()
        )
        ax.scatter(
            cross_comparison_group["R_diff", "mean"],
            cross_comparison_group[f"{player}_B-W", "mean"],
            s=cross_comparison_group[f"R ({player})", "percent"] * 1024,
            alpha=0.5,
        )
        ax.plot(
            cross_comparison_group["R_diff", "mean"],
            cross_comparison_group[f"{player}_B-W", "mean"],
            label=f"{player}, UM+={universal_metric:.4f}",
        )

    ax.legend(loc="lower center")
    ax.grid(linestyle="--")
    ax.set_title(f"{algoA} vs {algoB}")
    ax.set_xlabel(f"Difference of R ({algoA}) and R ({algoB})")
    ax.set_ylabel("B-W Metric")
    ax.set_xlim(-1, 1)
    ax.set_xticks(np.arange(-1.0, 1.1, 0.1))
    plt.tight_layout()
    plt.show()

    return result


if __name__ == "__main__":
    values = np.linspace(0, 1, 100)
    bins = 10
    plt.plot(values, [get_bin(value, bins) for value in values])
    plt.grid(True)
    plt.title(f"Equal-width binning with {bins} bins")
    plt.xlabel("Value")
    plt.ylabel("Bin")
    plt.show()