const { useState, useMemo } = React;

const OPTIMUM_COLOR = 'oklch(52% 0.17 145)';

const PALETTES = {
  classic: { train: 'oklch(55% 0.13 250)', val: 'oklch(70% 0.13 60)', optimum: OPTIMUM_COLOR },
  forest: { train: 'oklch(48% 0.12 165)', val: 'oklch(62% 0.14 35)', optimum: OPTIMUM_COLOR },
  mono: { train: 'oklch(30% 0.02 260)', val: 'oklch(60% 0.02 260)', optimum: OPTIMUM_COLOR },
  vivid: { train: 'oklch(55% 0.20 290)', val: 'oklch(70% 0.16 35)', optimum: OPTIMUM_COLOR }
};

function App() {
  const data = window.LOSS_DATA;
  const [tw, setTweak] = window.useTweaks({
    metric: 'loss',
    palette: 'classic'
  });
  const t = { ...tw, setTweak };

  const palette = PALETTES[t.palette] || PALETTES.classic;

  // Summary stats
  const stats = useMemo(() => {
    const last = data[data.length - 1];
    const first = data[0];
    const vals = data.map((d) => d.val);
    const minVal = Math.min(...vals);
    const minValIdx = vals.indexOf(minVal);
    const accs = data.map((d) => d.acc);
    const maxAcc = Math.max(...accs);
    const maxAccIdx = accs.indexOf(maxAcc);
    const dices = data.map((d) => d.dice);
    const maxDice = Math.max(...dices);
    // total time
    const totalSec = data.reduce((acc, d) => {
      const [hh, mm, ss] = d.time.split(':').map(Number);
      return acc + hh * 3600 + mm * 60 + ss;
    }, 0);
    const hours = Math.floor(totalSec / 3600);
    const mins = Math.floor(totalSec % 3600 / 60);
    const avgPer = totalSec / data.length;
    const avgMin = Math.floor(avgPer / 60);
    const avgSec = Math.round(avgPer % 60);
    // gap (computed at the optimum checkpoint — epoch 18, the one the model uses)
    const gapTrain = data[minValIdx].train;
    const gapVal = data[minValIdx].val;
    const gap = gapTrain - gapVal;
    return {
      first, last, minVal, minValIdx, maxAcc, maxAccIdx, maxDice,
      totalH: hours, totalM: mins,
      avgMin, avgSec,
      gap, gapTrain, gapVal,
      epochs: data.length
    };
  }, [data]);

  const fmt = window.lossFmt;

  const tabs = [
  { id: 'loss', label: 'Loss' },
  { id: 'acc', label: 'Accuracy' },
  { id: 'dice', label: 'Dice' }];


  return (
    <div className="page">
      <header className="topbar">
        <div className="brand">
          <div className="logo-mark">
            <svg viewBox="0 0 24 24" width="20" height="20">
              <path d="M3 21 L3 3 M3 21 L21 21" stroke="currentColor" strokeWidth="1.6" fill="none" strokeLinecap="round" />
              <path d="M5 17 C 8 15, 9 11, 11 9 S 16 6, 20 5" stroke={palette.train} strokeWidth="1.8" fill="none" strokeLinecap="round" />
              <path d="M5 18 C 8 14, 10 10, 13 9.5 S 18 9, 20 9" stroke={palette.val} strokeWidth="1.8" fill="none" strokeLinecap="round" />
            </svg>
          </div>
          <div className="brand-text">
            <div className="brand-name">plots<span className="brand-dot">·</span><span className="brand-sub">research</span></div>
          </div>
        </div>
        <div className="crumbs">
          <span>experiments</span>
          <span className="sep">/</span>
          <span>semantic-segmentation</span>
          <span className="sep">/</span>
          <span className="crumb-active">deeplab-v3 · 7-band · run 04</span>
        </div>
        <div className="topbar-r">
          <span className="status-pill"><span className="status-dot" />completed</span>
          <span className="kbd">⌘K</span>
        </div>
      </header>

      <window.LossSubNav accentColor="oklch(55% 0.13 250)" />

      <main className="grid">
        <section className="card chart-card" id="chart">
          <div className="card-hd">
            <div>
              <div className="eyebrow">Training run · 7-Band model</div>
              <h1 className="title">Καμπύλες απώλειας ανά εποχή</h1>
              <div className="subtitle">Training &amp; validation loss across {stats.epochs} epochs · DeepLab v3, 7 input bands</div>
            </div>
            <div className="meta-chips">
              <div className="chip"><span className="chip-k">epochs</span><span className="chip-v">{stats.epochs}</span></div>
              <div className="chip"><span className="chip-k">batches/epoch</span><span className="chip-v">~672</span></div>
              <div className="chip"><span className="chip-k">total time</span><span className="chip-v">{stats.totalH}h {stats.totalM}m</span></div>
            </div>
          </div>

          <div className="card-tools">
            <div className="tabs">
              {tabs.map((tab) =>
              <button
                key={tab.id}
                className={'tab' + (t.metric === tab.id ? ' active' : '')}
                onClick={() => t.setTweak('metric', tab.id)}>
                {tab.label}</button>
              )}
            </div>
          </div>

          <div className="legend">
            {t.metric === 'loss' &&
            <>
                <div className="legend-i"><span className="lg-line" style={{ background: palette.train }} />Training loss</div>
                <div className="legend-i"><span className="lg-line" style={{ background: palette.val }} />Validation loss</div>
              </>
            }
            {t.metric === 'acc' && <div className="legend-i"><span className="lg-line" style={{ background: palette.train }} />Validation accuracy</div>}
            {t.metric === 'dice' && <div className="legend-i"><span className="lg-line" style={{ background: palette.val }} />Validation Dice coefficient</div>}
            <div className="legend-i"><span className="lg-vline" style={{ background: palette.optimum }} />{t.metric === 'loss' ? 'Optimum epoch (min val loss)' : t.metric === 'acc' ? 'Optimum epoch (peak accuracy)' : 'Optimum epoch (peak Dice)'}</div>
            <div className="legend-i"><span className="lg-dot" style={{ background: t.metric === 'loss' || t.metric === 'acc' ? palette.train : palette.val, borderColor: 'var(--bg)' }} />Each point = one epoch</div>
          </div>

          <window.LossChart
            data={data}
            metric={t.metric}
            palette={palette} />
        </section>

        <aside className="card stats-card">
          <div className="stats-hd">Summary</div>
          <div className="stat-block">
            <div className="stat-l">Best validation loss</div>
            <div className="stat-v"><span className="big-num">{fmt(stats.minVal, 4)}</span><span className="stat-meta">epoch {data[stats.minValIdx].epoch}</span></div>
            <div className="stat-bar"><span style={{ width: `${(1 - stats.minVal / stats.first.val) * 100}%`, background: palette.val }} /></div>
            <div className="stat-sub">Δ from start: −{fmt((1 - stats.minVal / stats.first.val) * 100, 1)}% · runner-up epoch 16 ({fmt(data[16].val, 4)})</div>
          </div>

          <div className="stat-block">
            <div className="stat-l">Training loss @ optimum</div>
            <div className="stat-v"><span className="big-num">{fmt(data[stats.minValIdx].train, 4)}</span><span className="stat-meta">epoch {data[stats.minValIdx].epoch} (best checkpoint)</span></div>
            <div className="stat-bar"><span style={{ width: `${(1 - data[stats.minValIdx].train / stats.first.train) * 100}%`, background: palette.train }} /></div>
            <div className="stat-sub">Δ from start: −{fmt((1 - data[stats.minValIdx].train / stats.first.train) * 100, 1)}% · final epoch {stats.last.epoch}: {fmt(stats.last.train, 4)}</div>
          </div>

          <div className="stat-row-2">
            <div>
              <div className="stat-l">Accuracy @ optimum</div>
              <div className="stat-v-sm">{fmt(data[stats.minValIdx].acc, 4)}</div>
              <div className="stat-sub">at epoch {data[stats.minValIdx].epoch} (best checkpoint)</div>
              <div className="stat-sub dim-extra">peak {fmt(stats.maxAcc, 4)} · epoch {data[stats.maxAccIdx].epoch}</div>
            </div>
            <div>
              <div className="stat-l">Dice @ optimum</div>
              <div className="stat-v-sm">{fmt(data[stats.minValIdx].dice, 4)}</div>
              <div className="stat-sub">at epoch {data[stats.minValIdx].epoch} (best checkpoint)</div>
              <div className="stat-sub dim-extra">peak {fmt(stats.maxDice, 4)} · epoch {data[data.map((d) => d.dice).indexOf(stats.maxDice)].epoch}</div>
            </div>
          </div>

          <div className="divider" />

          <div className="train-val-gap">
            <div className="gap-hd">
              <div>
                <div className="gap-title">Train–val gap</div>
                <div className="gap-explain">Difference between training and validation loss at the optimum checkpoint (epoch {data[stats.minValIdx].epoch}). A small, stable gap means the model generalises; a large or growing gap signals overfitting.</div>
              </div>
              <span className="gap-v">{fmt(stats.gap, 3)}</span>
            </div>
            <div className="gap-track">
              <div className="gap-pt" style={{ left: `${stats.gapVal / 1.0 * 100}%`, background: palette.val }} />
              <div className="gap-pt" style={{ left: `${stats.gapTrain / 1.0 * 100}%`, background: palette.train }} />
              <div className="gap-bridge" style={{
                left: `${Math.min(stats.gapTrain, stats.gapVal) * 100}%`,
                width: `${Math.abs(stats.gap) * 100}%`
              }} />
            </div>
            <div className="gap-axis">
              <span>0</span>
              <span className="gap-axis-mark">val {fmt(stats.gapVal, 3)}</span>
              <span className="gap-axis-mark">train {fmt(stats.gapTrain, 3)}</span>
              <span>1.0</span>
            </div>
            <div className="stat-sub">Training loss sits {fmt(stats.gap, 3)} above validation at epoch {data[stats.minValIdx].epoch} — a sign of dropout / augmentation noise active during training but disabled at eval, rather than overfitting.</div>
          </div>

          <div className="divider" />

          <div className="runtime">
            <div className="rt-row">
              <span className="rt-k">Total wall time</span>
              <span className="rt-v">{stats.totalH}h {stats.totalM}m</span>
            </div>
            <div className="rt-row">
              <span className="rt-k">Avg per epoch</span>
              <span className="rt-v">{stats.avgMin}m {String(stats.avgSec).padStart(2, '0')}s</span>
            </div>
            <div className="rt-row">
              <span className="rt-k">Input bands</span>
              <span className="rt-v">7 bands</span>
            </div>
            <div className="rt-row">
              <span className="rt-k">Source</span>
              <span className="rt-v">training_validation_loss.json</span>
            </div>
          </div>
        </aside>

        <section className="card table-card" id="per-epoch">
          <div className="table-hd">
            <div className="table-title">Per-epoch metrics</div>
            <div className="table-sub">{stats.epochs} epochs · sortable values</div>
          </div>
          <div className="table-scroll">
            <table className="metrics-table">
              <thead>
                <tr>
                  <th>epoch</th>
                  <th>train loss</th>
                  <th>val loss</th>
                  <th>accuracy</th>
                  <th>Dice</th>
                  <th>time</th>
                  <th className="th-bar">progression</th>
                </tr>
              </thead>
              <tbody>
                {data.map((d) => {
                  const isMinVal = d.epoch === stats.minValIdx;
                  const isMaxAcc = d.epoch === stats.maxAccIdx;
                  return (
                    <tr key={d.epoch} className={isMinVal ? 'row-best' : ''}>
                      <td className="mono">{String(d.epoch).padStart(2, '0')}</td>
                      <td className="mono">{fmt(d.train)}</td>
                      <td className={'mono' + (isMinVal ? ' cell-hi' : '')}>
                        {fmt(d.val)}{isMinVal && <span className="badge-best">min</span>}
                      </td>
                      <td className={'mono' + (isMaxAcc ? ' cell-hi' : '')}>{fmt(d.acc)}</td>
                      <td className="mono">{fmt(d.dice)}</td>
                      <td className="mono dim">{d.time}</td>
                      <td className="td-bar">
                        <div className="mini-bars">
                          <div className="mini-bar-track">
                            <div className="mini-bar" style={{ width: `${d.train / 1.0 * 100}%`, background: palette.train }} />
                          </div>
                          <div className="mini-bar-track">
                            <div className="mini-bar" style={{ width: `${d.val / 1.0 * 100}%`, background: palette.val }} />
                          </div>
                        </div>
                      </td>
                    </tr>);

                })}
              </tbody>
            </table>
          </div>
        </section>

        <window.PerClassPanel accentColor={palette.train} />
      </main>

      <window.TweaksPanel title="Tweaks">
        <window.TweakSection title="Palette">
          <window.TweakSelect
            label="Color theme"
            value={t.palette}
            onChange={(v) => t.setTweak('palette', v)}
            options={[
            { value: 'classic', label: 'Classic (blue/amber)' },
            { value: 'forest', label: 'Forest (teal/copper)' },
            { value: 'mono', label: 'Monochrome' },
            { value: 'vivid', label: 'Vivid' }]
            } />
          
        </window.TweakSection>
      </window.TweaksPanel>
    </div>);

}

ReactDOM.createRoot(document.getElementById('root')).render(<App />);