File:Convergence of multinomial distribution to the gaussian distribution.webm

https://en.wikipedia.org/wiki/Multinomial_distribution#Large_deviation_theory for details of what this image shows.

```python import numpy as np import matplotlib.pyplot as plt from scipy.stats import multinomial from matplotlib.patches import RegularPolygon import os from tqdm import trange

M, N = 100000, 10 for N in trange(2, 200):

   p = np.array([0.2, 0.3, 0.5])

   samples = multinomial.rvs(N, p, size=M).T
   K = np.array([[-np.sqrt(1/2), np.sqrt(1/2), 0], [-np.sqrt(1/6), -np.sqrt(1/6), np.sqrt(4/6)]])
   result = np.dot(K, samples) / N

   triangle_vertices = np.array([K[:, 0], K[:, 1], K[:, 2]])

   def f(x, y):
       return -N/2 * np.sum((np.array([1/3, 1/3, 1/3]) + x * K[0,:] + y*K[1,:] - p)**2 / p, axis=0)

   x_values = np.linspace(-np.sqrt(1/2), np.sqrt(1/2), 50)
   y_values = np.linspace(-np.sqrt(1/6), np.sqrt(4/6), 50)
   X, Y = np.meshgrid(x_values, y_values)

   Z = np.zeros_like(X)
   for i in range(X.shape[0]):
       for j in range(X.shape[1]):
           Z[i, j] = f(X[i, j], Y[i, j])
           
   hexbin_x = result[0]
   hexbin_y = result[1]

   plt.figure(figsize=(10, 10 * np.sqrt(3)))
   plt.hexbin(hexbin_x, hexbin_y, gridsize=50, cmap='YlGnBu', extent=(min(result[0]), max(result[0]), min(result[1]), max(result[1])),
             bins='log', mincnt=1, alpha=0.7, edgecolors='gray', linewidths=0.1)

   # Overlay heatmap of function f within the equilateral triangle
   plt.imshow(Z, extent=(-np.sqrt(1/2), np.sqrt(1/2), -np.sqrt(1/6), np.sqrt(4/6)),
             origin='lower', cmap='coolwarm', alpha=0.5)

   # Plot equilateral triangle
   triangle = plt.Polygon(triangle_vertices, edgecolor='black', closed=True, fill=False)
   plt.gca().add_patch(triangle)

   plt.xlim(-np.sqrt(1/2), np.sqrt(1/2))
   plt.ylim(-np.sqrt(1/6), np.sqrt(4/6))

   plt.title(f"N={N}, p={p}")
   plt.gca().set_aspect('equal', adjustable='box')
   plt.axis('off')
   dir_path = f"./multinomial"
   if not os.path.exists(dir_path):
       os.makedirs(dir_path)

   plt.savefig(f"{dir_path}/{N:03d}.png",bbox_inches='tight')

   plt.close()

import imageio.v3 as iio import os from natsort import natsorted import moviepy.editor as mp

for dir_path in ["./multinomial"]:

   file_names = natsorted((fn for fn in os.listdir(dir_path) if fn.endswith('.png')))

   # Create a list of image files and set the frame rate
   images = []
   fps = 12

   # Iterate over the file names and append the images to the list
   for file_name in file_names:
       file_path = os.path.join(dir_path, file_name)
       images.append(iio.imread(file_path))

   filename = dir_path[2:]
   clip = mp.ImageSequenceClip(images, fps=fps)
   clip.write_videofile(f"{filename}.mp4")

!ffmpeg -i multinomial.mp4 -c:v libvpx-vp9 -b:v 0 -crf 10 -c:a libvorbis multinomial.webm