', unsafe_allow_html=True)
# Use a dynamic and informative title for the expander
with st.expander(f"Results for {filename}", expanded=True):
# --- START: STREAMLINED METRICS ---
# A single, powerful row for the most important results.
key_metric_cols = st.columns(3)
# Metric 1: The Prediction
key_metric_cols[0].metric("Prediction", predicted_class)
# Metric 2: The Confidence (with level in tooltip)
confidence_icon = (
"🟢"
if max_confidence >= 0.8
else "🟡" if max_confidence >= 0.6 else "🔴"
)
key_metric_cols[1].metric(
"Confidence",
f"{confidence_icon} {max_confidence:.1%}",
help=f"Confidence Level: {confidence_desc}",
)
# Metric 3: Ground Truth + Correctness (Combined)
if true_label_idx is not None:
is_correct = predicted_class == true_label_str
delta_text = "✅ Correct" if is_correct else "❌ Incorrect"
# Use delta_color="normal" to let the icon provide the visual cue
key_metric_cols[2].metric(
"Ground Truth",
true_label_str,
delta=delta_text,
delta_color="normal",
)
else:
key_metric_cols[2].metric("Ground Truth", "N/A")
st.divider()
# --- END: STREAMLINED METRICS ---
# --- START: CONSOLIDATED CONFIDENCE ANALYSIS ---
st.markdown("##### Probability Breakdown")
# This custom bullet bar logic remains as it is highly specific and valuable
def create_bullet_bar(probability, width=20, predicted=False):
filled_count = int(probability * width)
bar = "▤" * filled_count + "▢" * (width - filled_count)
percentage = f"{probability:.1%}"
pred_marker = "↩ Predicted" if predicted else ""
return f"{bar} {percentage} {pred_marker}"
if probs is not None:
stable_prob, weathered_prob = probs[0], probs[1]
else:
st.error(
"❌ Probability values are missing. Please check the inference process."
)
# Default values to prevent further errors
stable_prob, weathered_prob = 0.0, 0.0
is_stable_predicted, is_weathered_predicted = (
int(prediction) == 0
), (int(prediction) == 1)
st.markdown(
f"""
Stable (Unweathered)
{create_bullet_bar(stable_prob, predicted=is_stable_predicted)}
Weathered (Degraded)
{create_bullet_bar(weathered_prob, predicted=is_weathered_predicted)}
""",
unsafe_allow_html=True,
)
# --- END: CONSOLIDATED CONFIDENCE ANALYSIS ---
st.divider()
# --- START: CLEAN METADATA FOOTER ---
# Secondary info is now a clean, single-line caption
st.caption(
f"Analyzed with **{st.session_state.get('model_select', 'Unknown')}** in **{inference_time:.2f}s**."
)
# --- END: CLEAN METADATA FOOTER ---
st.markdown("
', unsafe_allow_html=True
)
with st.expander("🔧 Technical Details", expanded=False):
st.markdown(
"""
**Model Architecture:**
- Convolutional layers for feature extraction
- Residual connections for gradient flow
- Fully connected layers for classification
- Softmax activation for probability distribution
**Performance Metrics:**
- Accuracy: 94.8-96.2% on validation set
- F1-Score: 94.3-95.9% across classes
- Robust to spectral noise and baseline variations
**Data Processing:**
- Input: Raman spectra (any length)
- Resampling: Linear interpolation to 500 points
- Normalization: None (preserves intensity relationships)
"""
)
st.markdown(
"
", unsafe_allow_html=True
) # Close the wrapper div
render_time = time.time() - start_render
log_message(
f"col2 rendered in {render_time:.2f}s, active tab: {active_tab}"
)
with st.expander("Spectrum Preprocessing Results", expanded=False):
st.caption("