File size: 6,907 Bytes
b0bcfd5
 
e3566c9
b0bcfd5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e3566c9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b0bcfd5
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
{% extends "base.html" %}

{% block title %}Evaluation — AI Medical Intelligence Pipeline{% endblock %}

{% block content %}
<section class="hero">
  <div class="hero-text">
    <h1>Model Evaluation</h1>
    <p>
      Calibration metrics, confidence band analysis, and probability
      distribution from the inference pipeline.
    </p>
  </div>
</section>

<!-- Calibration metrics -->
{% if calib %}
<section class="eval-grid">
  <article class="panel">
    <h3>Calibration Parameters</h3>
    <div class="kv-group">
      <div class="kv">
        <span>Method</span><strong>{{ calib.get('method', 'N/A') }}</strong>
      </div>
      <div class="kv">
        <span>Temperature</span
        ><strong>{{ '%.4f'|format(calib.temperature) }}</strong>
      </div>
      <div class="kv">
        <span>Decision Threshold</span
        ><strong>{{ '%.4f'|format(calib.calibrated_threshold) }}</strong>
      </div>
      <div class="kv">
        <span>Base Threshold</span
        ><strong>{{ '%.4f'|format(calib.base_threshold) }}</strong>
      </div>
      <div class="kv">
        <span>High Band ≥</span><strong>{{ calib.high_threshold }}</strong>
      </div>
      <div class="kv">
        <span>Low Band &lt;</span><strong>{{ calib.low_threshold }}</strong>
      </div>
    </div>
  </article>

  <article class="panel">
    <h3>Calibration Quality</h3>
    <div class="metric-grid">
      <div class="metric-card">
        <div class="metric-label">ECE (Raw)</div>
        <div class="metric-value">{{ '%.4f'|format(calib.raw_ece) }}</div>
      </div>
      <div class="metric-card">
        <div class="metric-label">ECE (Calibrated)</div>
        <div class="metric-value">{{ '%.4f'|format(calib.cal_ece) }}</div>
      </div>
      <div class="metric-card">
        <div class="metric-label">Brier (Raw)</div>
        <div class="metric-value">{{ '%.4f'|format(calib.raw_brier) }}</div>
      </div>
      <div class="metric-card">
        <div class="metric-label">Brier (Cal)</div>
        <div class="metric-value">{{ '%.4f'|format(calib.cal_brier) }}</div>
      </div>
    </div>
    <p class="muted small" style="margin-top: 12px">
      Temperature scaling adjusts logits by T={{
      '%.4f'|format(calib.temperature) }} to produce better-calibrated
      probabilities. Lower ECE = better calibration.
    </p>
  </article>
</section>
{% endif %}

<!-- Normalization -->
{% if norm %}
<section class="panel" style="margin-top: 16px">
  <h3>Normalization Statistics</h3>
  <div class="kv-group" style="max-width: 500px">
    <div class="kv">
      <span>Mean (per channel)</span><strong>{{ norm.mean }}</strong>
    </div>
    <div class="kv">
      <span>Std (per channel)</span><strong>{{ norm.std }}</strong>
    </div>
    <div class="kv">
      <span>Computed from</span
      ><strong>{{ norm.get('n_images', 'N/A') }} images</strong>
    </div>
  </div>
</section>
{% endif %}

<!-- Confidence Band Breakdown -->
<section class="panel" style="margin-top: 16px">
  <h3>Confidence Band Analysis</h3>
  <p class="muted small">
    Distribution of {{ total }} processed cases across the three confidence
    bands.
  </p>

  <div class="band-grid">
    {% for bnd in ['HIGH', 'MEDIUM', 'LOW'] %} {% set d = band_data.get(bnd,
    {'total': 0, 'positive': 0, 'negative': 0}) %}
    <div class="band-card band-{{ bnd|lower }}">
      <div class="band-header">
        <span class="badge badge-{{ bnd|lower }}">{{ bnd }}</span>
        <span class="band-total">{{ d.total }} cases</span>
      </div>
      <div class="band-bars">
        <div class="band-bar-row">
          <span class="band-bar-label">Positive</span>
          <div class="band-bar">
            <div
              class="band-bar-fill fill-red"
              style="width: {{ (d.positive / d.total * 100) if d.total else 0 }}%"
            ></div>
          </div>
          <span class="band-bar-val">{{ d.positive }}</span>
        </div>
        <div class="band-bar-row">
          <span class="band-bar-label">Negative</span>
          <div class="band-bar">
            <div
              class="band-bar-fill fill-green"
              style="width: {{ (d.negative / d.total * 100) if d.total else 0 }}%"
            ></div>
          </div>
          <span class="band-bar-val">{{ d.negative }}</span>
        </div>
      </div>
    </div>
    {% endfor %}
  </div>
</section>

<!-- Probability Distribution -->
<section class="panel" style="margin-top: 16px">
  <h3>Calibrated Probability Distribution</h3>
  <p class="muted small">
    Histogram of calibrated probabilities across all cases (10 bins).
  </p>

  <div class="histogram">
    {% set max_bin = bins|max if bins|max > 0 else 1 %} {% for count in bins %}
    <div class="hist-col">
      <div
        class="hist-bar"
        style="height: {{ (count / max_bin * 180)|round }}px"
        title="{{ '%.1f'|format(loop.index0 * 0.1) }}–{{ '%.1f'|format(loop.index0 * 0.1 + 0.1) }}: {{ count }}"
      >
        <span class="hist-count">{{ count }}</span>
      </div>
      <div class="hist-label">{{ '%.1f'|format(loop.index0 * 0.1) }}</div>
    </div>
    {% endfor %}
  </div>
</section>

<!-- Summary stats -->
<section class="panel" style="margin-top: 16px">
  <h3>Summary Statistics</h3>
  <div class="kv-group" style="max-width: 500px">
    <div class="kv">
      <span>Total processed</span><strong>{{ stats.total }}</strong>
    </div>
    <div class="kv">
      <span>Positive (flagged)</span><strong>{{ stats.positive }}</strong>
    </div>
    <div class="kv">
      <span>Negative</span><strong>{{ stats.negative }}</strong>
    </div>
    <div class="kv">
      <span>Urgent escalations</span><strong>{{ stats.urgent }}</strong>
    </div>
    <div class="kv">
      <span>Average calibrated prob</span
      ><strong>{{ '%.4f'|format(stats.avg_cal_prob) }}</strong>
    </div>
    <div class="kv">
      <span>Heatmaps generated</span><strong>{{ stats.heatmaps }}</strong>
    </div>
  </div>
</section>

{% if gt_stats %}
<section class="panel" style="margin-top: 16px">
  <h3>Ground Truth Agreement</h3>
  {% if gt_stats.total == 0 %}
  <p class="muted small">No ground truth labels available yet.</p>
  {% else %}
  <div class="kv-group" style="max-width: 500px">
    <div class="kv">
      <span>Labeled Cases</span><strong>{{ gt_stats.total }}</strong>
    </div>
    <div class="kv">
      <span>Accuracy</span>
      <strong>{{ '%.1f'|format(gt_stats.accuracy * 100) }}%</strong>
    </div>
    <div class="kv">
      <span>False Positive Rate</span>
      <strong>{{ '%.1f'|format(gt_stats.fp_rate * 100) }}%</strong>
    </div>
    <div class="kv">
      <span>TP / TN</span><strong>{{ gt_stats.tp }} / {{ gt_stats.tn }}</strong>
    </div>
    <div class="kv">
      <span>FP / FN</span><strong>{{ gt_stats.fp }} / {{ gt_stats.fn }}</strong>
    </div>
  </div>
  {% endif %}
</section>
{% endif %}
{% endblock %}