File size: 5,748 Bytes
7845551
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
import gradio as gr
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd

from ppi.distributions import BinaryYFDistribution
from ppi.compute_curve import compute_curve


def convert_y_params_to_f_params(p_y, p_f_given_y1, p_f_given_y0):
    """Convert Y-based parameters to F-based parameters using Bayes' theorem"""
    # P(F = 1) = P(F = 1 | Y = 1) * P(Y = 1) + P(F = 1 | Y = 0) * P(Y = 0)
    p_f = p_f_given_y1 * p_y + p_f_given_y0 * (1 - p_y)

    # P(Y = 1 | F = 1) = P(F = 1 | Y = 1) * P(Y = 1) / P(F = 1)
    p_y_given_f1 = (p_f_given_y1 * p_y) / p_f if p_f > 0 else 0

    # P(Y = 1 | F = 0) = P(F = 0 | Y = 1) * P(Y = 1) / P(F = 0)
    p_y_given_f0 = ((1 - p_f_given_y1) * p_y) / (1 - p_f) if p_f < 1 else 0

    return p_f, p_y_given_f1, p_y_given_f0


def plot_custom_probabilities(p_f, p_y_given_f1, p_y_given_f0, show_ppi):
    plt.close()
    fig, ax = plt.subplots(1, 1, figsize=(8, 6))

    binary_dist = BinaryYFDistribution(
        p_f=p_f, p_y_given_f1=p_y_given_f1, p_y_given_f0=p_y_given_f0
    )
    n_range = np.arange(4, 200, 2)
    data = compute_curve(binary_dist, n_range)
    data = pd.DataFrame(data)
    corr = binary_dist.correlation()
    ax.set_ylim(0.5, 2)

    best_rel_perf = (
        binary_dist.variance_y()
        - binary_dist.covariance_f_y() ** 2 / binary_dist.variance_f()
    ) / binary_dist.variance_y()

    # Always show PPI++
    sns.lineplot(
        data,
        x="n",
        y="relative_var_cf",
        ax=ax,
        label="Relative Variance PPI++",
        linewidth=2,
    )

    # Conditionally show PPI
    if show_ppi:
        sns.lineplot(
            data,
            x="n",
            y="relative_var_ppi",
            ax=ax,
            label="Relative Variance PPI",
            linewidth=2,
            linestyle="dotted",
        )

    # Add horizontal line at y=1
    ax.axhline(
        y=1, color="k", linestyle="dotted", alpha=0.7, label="Baseline (Sample Mean)"
    )

    # Add horizontal line at y=1
    ax.axhline(
        y=best_rel_perf,
        color="b",
        linestyle="dotted",
        alpha=0.7,
        label="Asymptotic Performance",
    )

    # Find and mark crossing points
    cf_data = data["relative_var_cf"].values
    n_data = data["n"].values

    # Find where PPI++ crosses y=1
    cf_crossings = np.where(np.diff(np.sign(cf_data - 1)))[0]
    for crossing in cf_crossings:
        if crossing < len(n_data) - 1:
            ax.axvline(x=n_data[crossing], color="red", linestyle="--", alpha=0.7)
            ax.text(
                n_data[crossing],
                0.55,
                f"n={n_data[crossing]}",
                rotation=0,
                ha="center",
                va="bottom",
                color="red",
            )

    ax.set_xlabel("Sample Size (n)")
    ax.set_ylabel("Relative Variance")
    ax.set_title(
        f"Relative Variance Analysis (Correlation of Psuedo-Label: {corr:.2f})"
    )
    ax.legend()
    ax.grid(True, alpha=0.3)

    return fig


def plot_y_based_probabilities(p_y, p_f_given_y1, p_f_given_y0, show_ppi):
    """Plot using Y-based parameters by converting to F-based parameters"""
    # Convert Y-based parameters to F-based parameters
    p_f, p_y_given_f1, p_y_given_f0 = convert_y_params_to_f_params(
        p_y, p_f_given_y1, p_f_given_y0
    )

    # Use the existing plotting function with converted parameters
    return plot_custom_probabilities(p_f, p_y_given_f1, p_y_given_f0, show_ppi)


# Create interface for F-based parameters (original)
f_based_interface = gr.Interface(
    fn=plot_custom_probabilities,
    inputs=[
        gr.Slider(0.05, 0.95, value=0.50, step=0.05, label="P(F = 1)"),
        gr.Slider(0.05, 0.95, value=0.60, step=0.05, label="P(Y = 1 | F = 1)"),
        gr.Slider(0.05, 0.95, value=0.40, step=0.05, label="P(Y = 1 | F = 0)"),
        gr.Checkbox(value=True, label="Show PPI curve"),
    ],
    outputs=gr.Plot(label="PPI++ Analysis", format="png"),
    title="Example: Specify in terms of psuedo-label prevalence and Y | F distribution",
    description="""
    Analyze relative variance curves of PPI and PPI++ (with Cross-Fitting), as compared to using the empirical mean of Y.

    **Inputs:**
    - P(F = 1): Prior probability of the binary pseudo-label
    - P(Y = 1 | F = 1): Conditional probability of label given pseudo-label = 1
    - P(Y = 1 | F = 0): Conditional probability of label given pseudo-label = 0
    """,
    live=True,
    flagging_mode="never",
)

# Create interface for Y-based parameters
y_based_interface = gr.Interface(
    fn=plot_y_based_probabilities,
    inputs=[
        gr.Slider(0.05, 0.95, value=0.1, step=0.05, label="P(Y = 1)"),
        gr.Slider(0.05, 0.95, value=0.9, step=0.05, label="P(F = 1 | Y = 1)"),
        gr.Slider(0.05, 0.95, value=0.1, step=0.05, label="P(F = 1 | Y = 0)"),
        gr.Checkbox(value=True, label="Show PPI curve"),
    ],
    outputs=gr.Plot(label="PPI++ Analysis", format="png"),
    title="Example: Specify in terms of prevalance and F | Y distribution",
    description="""
    Analyze relative variance curves of PPI and PPI++ (with Cross-Fitting), as compared to using the empirical mean of Y.

    **Inputs:**
    - P(Y = 1): Prior probability of the binary label
    - P(F = 1 | Y = 1): Conditional probability of pseudo-label given label = 1
    - P(F = 1 | Y = 0): Conditional probability of pseudo-label given label = 0
    """,
    live=True,
    flagging_mode="never",
)

# Create tabbed interface
demo = gr.TabbedInterface(
    [f_based_interface, y_based_interface],
    ["F-based Parameters", "Y-based Parameters"],
    title="Sample Size Analysis (Binary Case)",
)


if __name__ == "__main__":
    demo.launch(share=True)