src/views/performance.py

import pandas as pd
import streamlit as st
import plotly.graph_objects as go
import plotly.express as px
from utils.metrics_data import load_metrics

st.markdown("## 📈 Performa Model")
st.caption("Evaluasi komprehensif model IndoBERT yang telah di-fine-tune pada dataset PRDECT-ID.")

metrics = load_metrics()
if metrics.get("is_demo"):
    st.warning(
        "Menampilkan data demo (hasil aktual dari Progress Proposal). "
        "Letakkan `metrics.json` hasil training di `model/metrics.json` untuk data real-time. "
        "Lihat `export_metrics_snippet.py`."
    )

cm = metrics["confusion_matrix"]

# ---- 4 kartu metrik ----
c1, c2, c3, c4 = st.columns(4)
card_specs = [
    ("ACCURACY", metrics["accuracy"], "Persentase prediksi benar dari seluruh data uji", c1),
    ("PRECISION", metrics["precision"], "Ketepatan prediksi positif dari semua prediksi positif", c2),
    ("RECALL", metrics["recall"], "Kemampuan menemukan semua sampel positif aktual", c3),
    ("F1-SCORE", metrics["f1"], "Harmonic mean dari Precision dan Recall", c4),
]
for label, val, desc, col in card_specs:
    with col:
        with st.container(border=True):
            st.caption(f"⭐ {label}")
            st.markdown(f"### {val*100:.1f}%")
            st.caption(desc)

st.write("")
col_cm, col_curve = st.columns(2)

with col_cm:
    with st.container(border=True):
        st.markdown(f"**Confusion Matrix**")
        st.caption(f"Test set · {metrics.get('n_test', cm['tn']+cm['fp']+cm['fn']+cm['tp'])} sampel")

        z = [[cm["tp"], cm["fn"]], [cm["fp"], cm["tn"]]]
        x_labels = ["Pred: Positive", "Pred: Negative"]
        y_labels = ["Actual: Positive", "Actual: Negative"]
        fig_cm = go.Figure(data=go.Heatmap(
            z=z, x=x_labels, y=y_labels,
            colorscale=[[0, "#eef2ff"], [1, "#4338ca"]],
            text=z, texttemplate="%{text}", textfont={"size": 20},
            showscale=False,
        ))
        fig_cm.update_layout(margin=dict(l=10, r=10, t=10, b=10), height=320)
        st.plotly_chart(fig_cm, use_container_width=True)

        m1, m2 = st.columns(2)
        m1.metric("True Positive + True Negative", cm["tp"] + cm["tn"])
        m2.metric("False Positive + False Negative", cm["fp"] + cm["fn"])

with col_curve:
    with st.container(border=True):
        st.markdown("**Kurva Pelatihan**")
        st.caption("Training & validation loss per epoch")

        epochs = list(range(1, len(metrics["train_loss"]) + 1))
        fig_curve = go.Figure()
        fig_curve.add_trace(go.Scatter(
            x=epochs, y=metrics["train_loss"], mode="lines+markers",
            name="Training Loss", line=dict(color="#4338ca"),
        ))
        fig_curve.add_trace(go.Scatter(
            x=epochs, y=metrics["val_loss"], mode="lines+markers",
            name="Validation Loss", line=dict(color="#e74c3c"),
        ))
        fig_curve.update_layout(
            margin=dict(l=10, r=10, t=10, b=10), height=280,
            xaxis_title="Epoch", yaxis_title="Loss",
            legend=dict(orientation="h", yanchor="bottom", y=1.02),
        )
        st.plotly_chart(fig_curve, use_container_width=True)

        bcol1, bcol2 = st.columns(2)
        bcol1.metric("Best Val Loss", f"{min(metrics['val_loss']):.3f}")
        bcol2.metric("Total Epoch", len(metrics["train_loss"]))

st.write("")
with st.expander("Lihat akurasi training vs validation per epoch"):
    df_acc = pd.DataFrame({
        "Epoch": list(range(1, len(metrics["train_acc"]) + 1)),
        "Train Accuracy": metrics["train_acc"],
        "Val Accuracy": metrics["val_acc"],
    })
    fig_acc = px.line(
        df_acc, x="Epoch", y=["Train Accuracy", "Val Accuracy"],
        markers=True, color_discrete_sequence=["#4338ca", "#e74c3c"],
    )
    fig_acc.update_layout(yaxis_title="Accuracy", legend_title="")
    st.plotly_chart(fig_acc, use_container_width=True)
    st.dataframe(df_acc, use_container_width=True, hide_index=True)