chore: initialize git repo, add matplotlib dep, extend config

- Add .gitignore for Python/data/models
- Add matplotlib>=3.8.0 for eval plots
- Add PretrainConfig, FinetuneConfig, BalabitAdapterConfig, EvalConfig dataclasses

tests/test_trainer.py

@@ -0,0 +1,121 @@
+"""Tests for Flow Matching training pipeline."""
+from __future__ import annotations
+
+import json
+import math
+from pathlib import Path
+
+import numpy as np
+import pytest
+
+from ai_mouse.trainer import load_and_prepare_data, train, _augment
+
+
+def _make_synthetic_trace(start, end, n_moves=30):
+    """Create a synthetic trace dict mimicking real JSONL format."""
+    sx, sy = start
+    ex, ey = end
+    dist = math.hypot(ex - sx, ey - sy)
+    angle = math.degrees(math.atan2(ey - sy, ex - sx))
+
+    events = []
+    for i in range(n_moves):
+        t_frac = i / (n_moves - 1)
+        x = int(sx + (ex - sx) * t_frac + np.random.normal(0, 3))
+        y = int(sy + (ey - sy) * t_frac + np.random.normal(0, 3))
+        t = int(t_frac * 500 + np.random.normal(0, 5))
+        events.append({"type": "move", "x": x, "y": y, "t": max(0, t)})
+
+    events.sort(key=lambda e: e["t"])
+    events[0]["t"] = 0
+
+    last_t = events[-1]["t"]
+    events.append({"type": "down", "x": ex, "y": ey, "t": last_t + 50})
+    events.append({"type": "up", "x": ex, "y": ey, "t": last_t + 130})
+
+    return {
+        "meta": {"start": [sx, sy], "end": [ex, ey], "dist": int(dist), "angle": round(angle, 1)},
+        "events": events,
+    }
+
+
+@pytest.fixture
+def synthetic_traces_file(tmp_path):
+    """Create a temp JSONL file with 25 synthetic traces."""
+    traces_path = tmp_path / "traces.jsonl"
+    rng = np.random.default_rng(42)
+    lines = []
+    for _ in range(25):
+        sx, sy = int(rng.integers(50, 750)), int(rng.integers(50, 750))
+        angle = rng.uniform(0, 2 * math.pi)
+        dist = int(rng.integers(100, 500))
+        ex = int(sx + dist * math.cos(angle))
+        ey = int(sy + dist * math.sin(angle))
+        ex = max(0, min(800, ex))
+        ey = max(0, min(600, ey))
+        trace = _make_synthetic_trace((sx, sy), (ex, ey))
+        lines.append(json.dumps(trace))
+    traces_path.write_text("\n".join(lines), encoding="utf-8")
+    return traces_path
+
+
+class TestLoadAndPrepare:
+    def test_returns_correct_shapes(self, synthetic_traces_file):
+        seq, cond, click_durs = load_and_prepare_data(synthetic_traces_file, seq_len=64)
+        assert seq.shape[1] == 64
+        assert seq.shape[2] == 3
+        assert cond.shape[1] == 3
+        assert len(seq) > 0
+
+    def test_forward_starts_near_zero(self, synthetic_traces_file):
+        seq, cond, _ = load_and_prepare_data(synthetic_traces_file, seq_len=64)
+        assert abs(seq[:, 0, 0].mean()) < 0.15
+
+    def test_forward_ends_near_one(self, synthetic_traces_file):
+        seq, cond, _ = load_and_prepare_data(synthetic_traces_file, seq_len=64)
+        assert abs(seq[:, -1, 0].mean() - 1.0) < 0.15
+
+
+class TestAugment:
+    def test_augmentation_multiplies_data(self, synthetic_traces_file):
+        seq, cond, _ = load_and_prepare_data(synthetic_traces_file, seq_len=64)
+        n_orig = len(seq)
+        seq_aug, cond_aug = _augment(seq, cond)
+        assert len(seq_aug) == n_orig * 6
+        assert len(cond_aug) == n_orig * 6
+
+
+class TestTrain:
+    def test_train_produces_model_files(self, synthetic_traces_file, tmp_path):
+        output_dir = tmp_path / "models"
+        train(
+            data_path=synthetic_traces_file,
+            output_dir=output_dir,
+            epochs=3,
+            batch_size=8,
+            seq_len=64,
+        )
+        assert (output_dir / "flow_model.pt").exists()
+        assert (output_dir / "click_dist.json").exists()
+        assert (output_dir / "duration_dist.json").exists()
+        assert (output_dir / "train_config.json").exists()
+
+    def test_train_loss_decreases(self, synthetic_traces_file, tmp_path):
+        output_dir = tmp_path / "models"
+        losses = []
+
+        def cb(msg):
+            if "loss" in msg:
+                losses.append(msg["loss"])
+
+        train(
+            data_path=synthetic_traces_file,
+            output_dir=output_dir,
+            epochs=20,
+            batch_size=8,
+            seq_len=64,
+            progress_callback=cb,
+        )
+        first_half = np.mean(losses[:10])
+        second_half = np.mean(losses[10:])
+        assert second_half < first_half