Create app.py

app.py

@@ -0,0 +1,189 @@
+# skew_normal_explorer_app.py
+# Run locally:
+#   pip install -r requirements.txt
+#   python skew_normal_explorer_app.py
+
+import numpy as np
+import gradio as gr
+import matplotlib.pyplot as plt
+from math import sqrt, pi
+from scipy.stats import skewnorm
+
+def theoretical_stats_skewnorm(alpha, mu, sigma):
+    """
+    Compute theoretical moments for the Skew-Normal(alpha, loc=mu, scale=sigma).
+    Returns mean, median, mode (numeric), variance, std dev, IQR, skewness,
+    kurtosis (Pearson) & excess kurtosis.
+    """
+    m, v, s, exk = skewnorm.stats(alpha, loc=mu, scale=sigma, moments="mvsk")
+    m = float(m); v = float(v); s = float(s); exk = float(exk)
+    std = float(np.sqrt(v))
+
+    # Quartiles
+    q1 = float(skewnorm.ppf(0.25, alpha, loc=mu, scale=sigma))
+    q3 = float(skewnorm.ppf(0.75, alpha, loc=mu, scale=sigma))
+    iqr = q3 - q1
+
+    # Median
+    median = float(skewnorm.ppf(0.5, alpha, loc=mu, scale=sigma))
+
+    # Mode via grid search
+    xs = np.linspace(m - 6*std, m + 6*std, 4001)
+    pdf_vals = skewnorm.pdf(xs, alpha, loc=mu, scale=sigma)
+    mode = float(xs[int(np.argmax(pdf_vals))])
+
+    return {
+        "mean": m,
+        "median": median,
+        "mode": mode,
+        "variance": v,
+        "std_dev": std,
+        "IQR": iqr,
+        "range": float("inf"),
+        "skewness": s,
+        "kurtosis": exk + 3.0,
+        "excess_kurtosis": exk
+    }
+
+def sample_stats(sample):
+    n = len(sample)
+    if n < 2:
+        s_mean = float(sample[0]) if n == 1 else float("nan")
+        return {
+            "mean": s_mean, "median": s_mean, "mode": s_mean,
+            "variance": 0.0, "std_dev": 0.0, "IQR": 0.0, "range": 0.0,
+            "skewness": 0.0, "kurtosis": 3.0, "excess_kurtosis": 0.0
+        }
+
+    s = np.asarray(sample, dtype=float)
+    s_mean = float(np.mean(s))
+    s_median = float(np.median(s))
+
+    # Mode estimate via histogram
+    counts, bin_edges = np.histogram(s, bins=min(50, max(5, int(np.sqrt(n)))))
+    max_bin_idx = int(np.argmax(counts))
+    mode_est = float((bin_edges[max_bin_idx] + bin_edges[max_bin_idx+1]) / 2.0)
+
+    s_var = float(np.var(s, ddof=1))
+    s_std = float(np.sqrt(s_var))
+
+    q1, q3 = np.percentile(s, [25, 75])
+    iqr = q3 - q1
+    s_range = float(np.max(s) - np.min(s))
+
+    m2 = np.mean((s - s_mean)**2)
+    m3 = np.mean((s - s_mean)**3)
+    m4 = np.mean((s - s_mean)**4)
+    if m2 <= 0:
+        skew, kurt = 0.0, 3.0
+    else:
+        skew = m3 / (m2 ** 1.5)
+        kurt = m4 / (m2 ** 2)
+    ex_kurt = kurt - 3.0
+
+    return {
+        "mean": s_mean,
+        "median": s_median,
+        "mode": mode_est,
+        "variance": s_var,
+        "std_dev": s_std,
+        "IQR": iqr,
+        "range": s_range,
+        "skewness": skew,
+        "kurtosis": kurt,
+        "excess_kurtosis": ex_kurt
+    }
+
+def format_stats_block(title, d):
+    range_str = "∞" if d["range"] == float("inf") else f"{d['range']:.6g}"
+    lines = [
+        f"**{title}**",
+        f"- Mean: {d['mean']:.6g}",
+        f"- Median: {d['median']:.6g}",
+        f"- Mode: {d['mode']:.6g}",
+        f"- Variance: {d['variance']:.6g}",
+        f"- Std Dev: {d['std_dev']:.6g}",
+        f"- IQR: {d['IQR']:.6g}",
+        f"- Range: {range_str}",
+        f"- Skewness: {d['skewness']:.6g}",
+        f"- Kurtosis: {d['kurtosis']:.6g}",
+        f"- Excess Kurtosis: {d['excess_kurtosis']:.6g}",
+    ]
+    return "\n".join(lines)
+
+def render(alpha, mu, sigma, n, seed, x_min, x_max, bins, show_hist, overlay_empirical_pdf):
+    sigma = max(1e-6, sigma)
+
+    if x_min >= x_max:
+        theo_tmp = theoretical_stats_skewnorm(alpha, mu, sigma)
+        m, std = theo_tmp["mean"], max(1e-9, theo_tmp["std_dev"])
+        x_min, x_max = m - 4*std, m + 4*std
+
+    x = np.linspace(x_min, x_max, 800)
+    y = skewnorm.pdf(x, alpha, loc=mu, scale=sigma)
+
+    rng = np.random.default_rng(int(seed))
+    sample = skewnorm.rvs(alpha, loc=mu, scale=sigma, size=int(n), random_state=rng)
+
+    theo = theoretical_stats_skewnorm(alpha, mu, sigma)
+    samp = sample_stats(sample)
+
+    fig, ax = plt.subplots(figsize=(8, 4.5), dpi=120)
+    ax.plot(x, y, label="Theoretical PDF (Skew-Normal)")
+    if show_hist:
+        ax.hist(sample, bins=int(bins), density=True, alpha=0.5, label="Sample histogram")
+
+    if overlay_empirical_pdf:
+        bw = 1.06 * max(1e-8, samp["std_dev"]) * (len(sample) ** (-1/5))
+        bw = max(bw, 1e-6)
+        diffs = (x.reshape(-1, 1) - sample.reshape(1, -1)) / bw
+        kernel_vals = np.exp(-0.5 * diffs**2) / (sqrt(2*pi) * bw)
+        kde = np.mean(kernel_vals, axis=1)
+        ax.plot(x, kde, linestyle="--", label="Empirical density (KDE-like)")
+
+    ax.set_title("Skew-Normal & Normal Explorer (α=0 gives Normal)")
+    ax.set_xlabel("x")
+    ax.set_ylabel("density")
+    ax.legend(loc="best")
+    ax.grid(True, linestyle="--")
+
+    left = format_stats_block("Theoretical (Skew-Normal)", theo)
+    right = format_stats_block("Sample (from sliders)", samp)
+    stats_md = left + "\n\n" + right
+
+    return fig, stats_md
+
+with gr.Blocks(title="Skew-Normal & Normal Explorer") as demo:
+    gr.Markdown("# Skew-Normal & Normal Explorer")
+    gr.Markdown(
+        "Slide **α (skewness)** to skew left/right. **α=0 → Normal(μ, σ²)**. "
+        "Adjust μ, σ, n, and window. Compare theoretical vs sample stats."
+    )
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            alpha = gr.Slider(-15.0, 15.0, value=0.0, step=0.1, label="Skewness (α)")
+            mu = gr.Slider(-10.0, 10.0, value=0.0, step=0.1, label="Location (μ)")
+            sigma = gr.Slider(0.1, 10.0, value=1.0, step=0.1, label="Scale (σ)")
+            n = gr.Slider(10, 200000, value=2000, step=10, label="Sample size (n)")
+            seed = gr.Slider(0, 99999, value=42, step=1, label="Random seed")
+
+            with gr.Accordion("Plot window & layers", open=False):
+                x_min = gr.Number(value=-5.0, label="x min")
+                x_max = gr.Number(value=5.0, label="x max")
+                bins = gr.Slider(5, 200, value=40, step=1, label="Histogram bins")
+                show_hist = gr.Checkbox(value=True, label="Show sample histogram")
+                overlay_empirical_pdf = gr.Checkbox(value=False, label="Overlay empirical density (KDE-like)")
+
+        with gr.Column(scale=2):
+            plot = gr.Plot(label="Curve / Histogram")
+            stats = gr.Markdown(label="Descriptive Statistics")
+
+    inputs = [alpha, mu, sigma, n, seed, x_min, x_max, bins, show_hist, overlay_empirical_pdf]
+    demo.load(render, inputs=inputs, outputs=[plot, stats])
+    for w in inputs:
+        w.change(render, inputs=inputs, outputs=[plot, stats])
+
+if __name__ == "__main__":
+    demo.launch()
+