File size: 2,109 Bytes
a955673 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 | import numpy as np
import matplotlib.pyplot as plt
import gradio as gr
def f(x, func_name="Quadratic"):
if func_name == "Quadratic":
return (x - 2)**2 + 1
elif func_name == "Quartic":
return x**4 - 3*(x**2) + 2
def grad_f(x, func_name="Quadratic"):
if func_name == "Quadratic":
return 2*(x - 2)
elif func_name == "Quartic":
return 4*(x**3) - 6*x
def run_gd(func_name, x0, lr, steps, x_min, x_max):
xs = [float(x0)]
ys = [float(f(x0, func_name))]
x = float(x0)
for _ in range(int(steps)):
g = float(grad_f(x, func_name))
x = x - float(lr) * g
xs.append(x)
ys.append(float(f(x, func_name)))
grid = np.linspace(float(x_min), float(x_max), 400)
vals = f(grid, func_name)
fig1 = plt.figure()
plt.plot(grid, vals)
plt.scatter(xs, ys, s=30)
plt.plot(xs, ys, linestyle="--")
plt.title(f"Gradient Descent Path on {func_name}")
plt.xlabel("x")
plt.ylabel("f(x)")
plt.grid(True)
fig2 = plt.figure()
plt.plot(range(len(ys)), ys)
plt.title("Objective Value Over Iterations")
plt.xlabel("iteration")
plt.ylabel("f(x)")
plt.grid(True)
final = f"Final x = {xs[-1]:.6f}, f(x) = {ys[-1]:.6f}"
return fig1, fig2, final
demo = gr.Interface(
fn=run_gd,
inputs=[
gr.Dropdown(["Quadratic", "Quartic"], value="Quadratic", label="Function"),
gr.Slider(-10, 10, value=8, step=0.1, label="Initial x0"),
gr.Slider(0.001, 1.0, value=0.1, step=0.001, label="Learning rate (lr)"),
gr.Slider(1, 200, value=30, step=1, label="Steps"),
gr.Slider(-15, 0, value=-5, step=0.5, label="Plot x_min"),
gr.Slider(0, 15, value=10, step=0.5, label="Plot x_max"),
],
outputs=[
gr.Plot(label="Function + GD path"),
gr.Plot(label="Loss curve"),
gr.Textbox(label="Result"),
],
title="Gradient Descent Visualizer (from scratch)",
description="Adjust learning rate, starting point, and steps to see how gradient descent moves. Update rule is implemented manually."
)
demo.launch()
|