File size: 6,814 Bytes
ff3bc98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
474491f
 
 
 
 
 
 
ff3bc98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib

matplotlib.use("Agg")


def load_evaluation_data(data_path) -> pd.DataFrame:
    """Load evaluation results from CSV"""
    eval_file = data_path / "evaluation_results.csv"
    if eval_file.exists():
        return pd.read_csv(eval_file)
    return None


def get_model_style(model_name):
    """
    Get color and hatch pattern for a model

    Color scheme:
    - GPT models: Gray (#808080)
    - Qwen2.5-VL-32B: Light purple (#9B87E8) - BASE solid, SFT with pattern
    - Qwen2-VL-72B: Medium blue (#5B7FD8) - BASE solid, SFT with pattern

    Returns:
        tuple: (color, hatch_pattern)
    """
    if "GPT" in model_name or "gpt" in model_name:
        return "#808080", None

    if "Qwen2.5" in model_name or "qwen2p5" in model_name or "32B" in model_name:
        if "SFT" in model_name:
            return "#9B87E8", "///"
        else:
            return "#9B87E8", None

    if "Qwen2" in model_name or "72B" in model_name:
        if "SFT" in model_name:
            return "#5B7FD8", "///"
        else:
            return "#5B7FD8", None

    if "Qwen3" in model_name or "qwen3" in model_name or "32B" in model_name:
        if "SFT" in model_name:
            return "#6B4DB8", "///"
        else:
            return "#6B4DB8", None

    return "#6B4DC1", None


def create_accuracy_plot(
    eval_df: pd.DataFrame,
    selected_models: list = None,
    selected_categories: list = None,
):
    """
    Create bar chart of accuracy by category, colored by model

    Args:
        eval_df: DataFrame with evaluation results
        selected_models: List of models to display (None for all)
        selected_categories: List of categories to display (None for all)

    Returns:
        matplotlib figure
    """
    if eval_df is None:
        return None

    # Filter data
    df_filtered = eval_df.copy()
    if selected_models:
        df_filtered = df_filtered[df_filtered["model"].isin(selected_models)]
    if selected_categories:
        df_filtered = df_filtered[df_filtered["category"].isin(selected_categories)]

    # Create figure
    fig, ax = plt.subplots(figsize=(12, 6))

    # Get unique categories and models
    categories = df_filtered["category"].unique()
    models = df_filtered["model"].unique()

    # Set up bar positions
    x = range(len(categories))
    width = 0.8 / len(models)

    for i, model in enumerate(models):
        model_data = df_filtered[df_filtered["model"] == model]
        accuracies = [
            model_data[model_data["category"] == cat]["accuracy"].values[0]
            for cat in categories
        ]

        color, hatch = get_model_style(model)

        offset = (i - len(models) / 2) * width + width / 2
        ax.bar(
            [xi + offset for xi in x],
            accuracies,
            width,
            label=model,
            color=color,
            hatch=hatch,
            alpha=0.8,
            edgecolor="white",
            linewidth=1.2,
        )

    # Customize plot
    ax.set_xlabel("Category", fontsize=12, fontweight="bold")
    ax.set_ylabel("Accuracy", fontsize=12, fontweight="bold")
    ax.set_title("Model Accuracy by Category", fontsize=14, fontweight="bold")
    ax.set_xticks(x)
    ax.set_xticklabels(categories, rotation=0)
    ax.set_ylim(0, 1.0)
    ax.legend(loc="lower right", framealpha=0.9)
    ax.grid(axis="y", alpha=0.3, linestyle="--")

    plt.tight_layout()
    return fig


def create_precision_recall_plot(
    eval_df: pd.DataFrame,
    selected_models: list = None,
    selected_categories: list = None,
):
    """
    Create subplot with precision and recall by category, colored by model

    Args:
        eval_df: DataFrame with evaluation results
        selected_models: List of models to display (None for all)
        selected_categories: List of categories to display (None for all)

    Returns:
        matplotlib figure
    """
    if eval_df is None:
        return None

    # Filter data
    df_filtered = eval_df.copy()
    if selected_models:
        df_filtered = df_filtered[df_filtered["model"].isin(selected_models)]
    if selected_categories:
        df_filtered = df_filtered[df_filtered["category"].isin(selected_categories)]

    # Create figure with subplots
    fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 6))

    # Get unique categories and models
    categories = df_filtered["category"].unique()
    models = df_filtered["model"].unique()

    # Set up bar positions
    x = range(len(categories))
    width = 0.8 / len(models)

    # Plot precision bars
    for i, model in enumerate(models):
        model_data = df_filtered[df_filtered["model"] == model]
        precisions = [
            model_data[model_data["category"] == cat]["precision"].values[0]
            for cat in categories
        ]

        # Get color and pattern for this model
        color, hatch = get_model_style(model)

        offset = (i - len(models) / 2) * width + width / 2
        ax1.bar(
            [xi + offset for xi in x],
            precisions,
            width,
            label=model,
            color=color,
            hatch=hatch,
            alpha=0.8,
            edgecolor="white",
            linewidth=1.2,
        )

    # Customize precision plot
    ax1.set_xlabel("Category", fontsize=12, fontweight="bold")
    ax1.set_ylabel("Precision", fontsize=12, fontweight="bold")
    ax1.set_title("Model Precision by Category", fontsize=14, fontweight="bold")
    ax1.set_xticks(x)
    ax1.set_xticklabels(categories, rotation=0)
    ax1.set_ylim(0, 1.0)
    ax1.legend(loc="lower right", framealpha=0.9)
    ax1.grid(axis="y", alpha=0.3, linestyle="--")

    # Plot recall bars
    for i, model in enumerate(models):
        model_data = df_filtered[df_filtered["model"] == model]
        recalls = [
            model_data[model_data["category"] == cat]["recall"].values[0]
            for cat in categories
        ]

        # Get color and pattern for this model
        color, hatch = get_model_style(model)

        offset = (i - len(models) / 2) * width + width / 2
        ax2.bar(
            [xi + offset for xi in x],
            recalls,
            width,
            label=model,
            color=color,
            hatch=hatch,
            alpha=0.8,
            edgecolor="white",
            linewidth=1.2,
        )

    ax2.set_xlabel("Category", fontsize=12, fontweight="bold")
    ax2.set_ylabel("Recall", fontsize=12, fontweight="bold")
    ax2.set_title("Model Recall by Category", fontsize=14, fontweight="bold")
    ax2.set_xticks(x)
    ax2.set_xticklabels(categories, rotation=0)
    ax2.set_ylim(0, 1.0)
    ax2.legend(loc="lower right", framealpha=0.9)
    ax2.grid(axis="y", alpha=0.3, linestyle="--")

    plt.tight_layout()
    return fig