TABCF/feature_usage_plot.py at main · ningzuow/TABCF · 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
import os
import json
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.patches import Patch

# Configuration to use LaTeX in Matplotlib
plt.rc('text', usetex=True)

# Define font sizes for different plot elements
title_fontsize = 20
xlabel_fontsize = 15
ylabel_fontsize = 40
tick_params_fontsize_x = 15  # Font size for x-axis ticks
tick_params_fontsize_y = 15  # Font size for y-axis ticks
bar_label_fontsize = 15
legend_fontsize = 15

dataname = "adult"
output_folder_name = f'counterfactual_results/{dataname}'

methods = ["Wachter", "DiCE", "Revise", "CCHVAE", "DiCE_latent"]

us = r'\textbf{T\Large{\textbf{AB}}\LARGE{C}\Large{\textbf{F}}}'
# Simplified and corrected LaTeX formatted names
names = [
    r'\textbf{Wachter}',
    r'\textbf{DiCE}',
    r'\textbf{REVISE}',
    r'\textbf{CCHVAE}',
    us
]

columns = ["marital.status", "education", "occupation", "age", "relationship", "sex", "hours.per.week", "workclass", "capital.gain", "capital.loss", "race", "native.country"]

# List of numerical columns
numerical_columns = ["age", "hours.per.week", "capital.gain", "capital.loss"]

columns.reverse()

fig, axs = plt.subplots(1, len(methods), figsize=(4 * len(methods), 6), sharey=True)

if len(methods) == 1:
    axs = [axs]

for i, method in enumerate(methods):
    output_folder_name_method = f'{output_folder_name}/{method}'

    if method == "DiCE_latent":
        output_folder_name_method = f'{output_folder_name_method}/num_encoding[min_max_torch]_num_loss[L2]_cat_actv[gumbel_softmax, tau=1.0]_cont_actv[sigmoid]_reparam/gradient/proximity_weight_input[1.0]_hidden_dims[16]_proximity_weight_latent[1.0]_proximity_latent_loss[L2]'
    if method == "DiCE":
        output_folder_name_method = f'{output_folder_name_method}/gradient/proximity_weight_input[1.0]_hidden_dims[16]'

    feature_c_json = f'{output_folder_name_method}/1000_samples/feature_changes.json'
    with open(feature_c_json, 'r') as file:
        data = pd.DataFrame.from_dict([json.load(file)])

    # Extract values for plotting
    values = data[columns].iloc[0]

    # Ensure values is a Series or 1D array
    if values.ndim > 1:
        values = values.squeeze()

    ax = axs[i]
    colors = ['royalblue' if col in numerical_columns else 'grey' for col in columns]
    bars = ax.barh(columns, values, color=colors)
    ax.set_title(names[i], fontsize=title_fontsize)

    # Calculate the total percentage for numerical and categorical features
    total_numerical = sum(values[col] for col in numerical_columns)
    total_categorical = sum(values[col] for col in columns if col not in numerical_columns)

    # Annotate each bar with its value inside the bar
    for bar in bars:
        width = bar.get_width()
        if width > 0.03:
            ax.text(
                width - (width * 0.02),  # Position the text slightly left from the end of the bar
                bar.get_y() + bar.get_height() / 2,
                f'{width:.2f}',  # Format the value with two decimal places
                va='center',
                ha='right',
                fontsize=bar_label_fontsize,
                color='white'
            )

    # Create legend patches
    legend_patches = [
        Patch(color='royalblue', label=f'Numerical: {total_numerical:.2f}%'),
        Patch(color='grey', label=f'Categorical: {total_categorical:.2f}%')
    ]

    # Add legend with total percentages for numerical and categorical features
    legend = ax.legend(
        handles=legend_patches,
        title='',
        loc='lower right',
        fontsize=legend_fontsize,
        frameon=True,  # Add box around the legend
        title_fontsize=legend_fontsize
    )

    # Customize tick parameters
    ax.tick_params(axis='x', which='major', labelsize=tick_params_fontsize_x)
    ax.tick_params(axis='y', which='major', labelsize=tick_params_fontsize_y)

    # Customize y-tick labels for numerical columns to be bold
    yticks = ax.get_yticks()
    yticklabels = [col if col not in numerical_columns else rf'\textbf{{{col}}}' for col in columns]

    yticklabels = [label.replace('hours.per.week', 'hours/week') for label in yticklabels]
    yticklabels = [label.replace('marital.status', 'marital\nstatus') for label in yticklabels]
    yticklabels = [label.replace('native.country', 'native\ncountry') for label in yticklabels]

    ax.set_yticks(yticks)
    ax.set_yticklabels(yticklabels, fontsize=tick_params_fontsize_y)

# Add an overall x-axis label
fig.text(0.5, 0.04, 'Feature utilization (\%)', ha='center', va='center', fontsize=title_fontsize)

plt.savefig(f'feature_changes_adult.png', format='png', dpi=300, bbox_inches='tight')