chore(repo): initialize reproducible research workspace

2026-09-20 05:17:11 +12:00
commit 1e7cc2a71b
36 changed files with 9169 additions and 0 deletions
@@ -0,0 +1,145 @@
+import importlib
+import json
+import os
+from pathlib import Path
+import sys
+import traceback
+
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+CACHE_ROOT = PROJECT_ROOT / ".cache"
+cache_settings = {
+    "LOCALAPPDATA": CACHE_ROOT / "localappdata",
+    "MPLCONFIGDIR": CACHE_ROOT / "matplotlib",
+    "NUMBA_CACHE_DIR": CACHE_ROOT / "numba",
+    "SKB_DATA_DIRECTORY": CACHE_ROOT / "skrub",
+    "HF_HOME": CACHE_ROOT / "huggingface",
+    "TORCH_HOME": CACHE_ROOT / "torch",
+    "XDG_CACHE_HOME": CACHE_ROOT,
+}
+for key, path in cache_settings.items():
+    path.mkdir(parents=True, exist_ok=True)
+    os.environ[key] = str(path)
+os.environ["PYTENSOR_FLAGS"] = f"base_compiledir={CACHE_ROOT / 'pytensor'}"
+
+import numpy as np
+
+
+results = {"python": sys.version, "imports": {}, "tests": {}}
+
+
+def record(name, fn):
+    try:
+        value = fn()
+        results["tests"][name] = {"status": "passed", "detail": value}
+    except Exception as exc:
+        results["tests"][name] = {
+            "status": "failed",
+            "detail": f"{type(exc).__name__}: {exc}",
+            "traceback": traceback.format_exc(),
+        }
+
+
+for module_name in [
+    "numpy", "scipy", "pandas", "polars", "pyarrow", "duckdb",
+    "statsmodels", "pymc", "pytensor", "arviz", "nutpie",
+    "jax", "numpyro", "blackjax", "bambi", "semopy",
+    "factor_analyzer", "pingouin", "girth", "torch", "torchvision",
+    "transformers", "sentence_transformers", "xgboost",
+    "tabpfn",
+]:
+    try:
+        module = importlib.import_module(module_name)
+        results["imports"][module_name] = {
+            "status": "passed",
+            "version": getattr(module, "__version__", "unknown"),
+        }
+    except Exception as exc:
+        results["imports"][module_name] = {
+            "status": "failed",
+            "detail": f"{type(exc).__name__}: {exc}",
+        }
+
+
+def pymc_test():
+    import pymc as pm
+
+    rng = np.random.default_rng(20260920)
+    observed = rng.normal(0.35, 1.0, size=40)
+    with pm.Model():
+        mu = pm.Normal("mu", 0.0, 1.0)
+        pm.Normal("y", mu, 1.0, observed=observed)
+        inference = pm.sample(
+            draws=50,
+            tune=50,
+            chains=1,
+            cores=1,
+            random_seed=20260920,
+            progressbar=False,
+            compute_convergence_checks=False,
+        )
+    posterior_mean = float(inference.posterior["mu"].mean())
+    return {"posterior_mean": posterior_mean, "draws": 50, "chains": 1}
+
+
+def sem_test():
+    import pandas as pd
+    import semopy
+
+    rng = np.random.default_rng(20260920)
+    latent = rng.normal(size=160)
+    frame = pd.DataFrame(
+        {
+            "x1": 0.8 * latent + rng.normal(scale=0.4, size=160),
+            "x2": 0.7 * latent + rng.normal(scale=0.5, size=160),
+            "x3": 0.9 * latent + rng.normal(scale=0.3, size=160),
+        }
+    )
+    model = semopy.Model("factor =~ x1 + x2 + x3")
+    fit = model.fit(frame)
+    return {"success": bool(fit.success), "objective": float(fit.fun)}
+
+
+def factor_test():
+    from factor_analyzer import FactorAnalyzer
+
+    rng = np.random.default_rng(20260920)
+    latent = rng.normal(size=(180, 1))
+    data = latent @ np.array([[0.8, 0.7, 0.9, 0.6]]) + rng.normal(
+        scale=0.45, size=(180, 4)
+    )
+    fitted = FactorAnalyzer(n_factors=1, rotation=None).fit(data)
+    return {"loading_shape": list(fitted.loadings_.shape)}
+
+
+def torch_test():
+    import torch
+
+    if not torch.cuda.is_available():
+        raise RuntimeError("torch.cuda.is_available() is false")
+    left = torch.randn((256, 256), device="cuda")
+    right = torch.randn((256, 256), device="cuda")
+    result = left @ right
+    torch.cuda.synchronize()
+    return {
+        "torch": torch.__version__,
+        "cuda_runtime": torch.version.cuda,
+        "device": torch.cuda.get_device_name(0),
+        "finite": bool(torch.isfinite(result).all().item()),
+    }
+
+
+record("pymc_sampling", pymc_test)
+record("semopy_cfa", sem_test)
+record("factor_analyzer", factor_test)
+record("torch_cuda", torch_test)
+
+serialized = json.dumps(results, ensure_ascii=False, indent=2)
+(Path(__file__).parent / "smoke-test-result.json").write_text(
+    serialized, encoding="utf-8"
+)
+print(serialized)
+
+failed_imports = [k for k, v in results["imports"].items() if v["status"] != "passed"]
+failed_tests = [k for k, v in results["tests"].items() if v["status"] != "passed"]
+raise SystemExit(1 if failed_imports or failed_tests else 0)