import gpustat, humanize

def _get_gpu_info():
    try:
        g = gpustat.GPUStatCollection.new_query()
        g = g.jsonify()["gpus"]
        name = g[0]["name"]; mem_used = g[0]["memory.used"]; mem_total = g[0]["memory.total"]
        return name, mem_used*1024**2, mem_total*1024**2
    except Exception:
        return "N/A", None, None

def env_report() -> Dict[str, Any]:
    gpu_name, mem_used_b, mem_total_b = _get_gpu_info()
    rep = {
        "python": platform.python_version(),
        "os": platform.platform(),
        "cpu_count": psutil.cpu_count(),
        "ram_total": humanize.naturalsize(psutil.virtual_memory().total, binary=True),
        "gpu_name": gpu_name,
        "torch": torch.__version__,
        "cuda_available": torch.cuda.is_available(),
        "cudnn": torch.backends.cudnn.version() if torch.cuda.is_available() else None,
        "numpy": np.__version__,
        "pandas": pd.__version__,
        "sklearn": __import__("sklearn").__version__,
        "rdkit": __import__("rdkit").__version__,
        "transformers": __import__("transformers").__version__,
        "econml": __import__("econml").__version__,
        "dowhy": __import__("dowhy").__version__,
    }
    if mem_total_b:
        rep["vram_used"]  = humanize.naturalsize(mem_used_b, binary=True)
        rep["vram_total"] = humanize.naturalsize(mem_total_b, binary=True)
    return rep

rep = env_report()
print(json.dumps(rep, indent=2))


# Clean, readable figures (no seaborn—stays portable)
mpl.rcParams.update({
    "figure.figsize": (5.5, 4.2),
    "axes.titlesize": 12,
    "axes.labelsize": 11,
    "xtick.labelsize": 9,
    "ytick.labelsize": 9,
    "legend.fontsize": 9,
    "figure.dpi": 160,
    "savefig.dpi": 300,
    "axes.spines.top": False,
    "axes.spines.right": False,
    "font.family": "DejaVu Sans"
})
def savefig(path: Path, bbox=True):
    path.parent.mkdir(parents=True, exist_ok=True)
    plt.tight_layout()
    plt.savefig(path, bbox_inches="tight" if bbox else None)


import contextlib, time

def save_json(obj, path: Path):
    path = Path(path); path.parent.mkdir(parents=True, exist_ok=True)
    with open(path, "w") as f: json.dump(obj, f, indent=2)

@contextlib.contextmanager
def timer(msg:str):
    t0 = time.time()
    print(f"⏱️ {msg} ...", end="", flush=True)
    yield
    dt = time.time() - t0
    print(f" done in {dt:.2f}s")

def peek(df: pd.DataFrame, n:int=5):
    display(df.head(n))
    print(f"shape={df.shape} | cols={list(df.columns)[:6]}{'...' if df.shape[1]>6 else ''}")

def require_schema(df: pd.DataFrame, required_cols):
    missing = [c for c in required_cols if c not in df.columns]
    if missing:
        raise ValueError(f"Missing required columns: {missing}")