animate_gd.py

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

from matplotlib import animation, rc
from IPython.display import HTML

# set ffmpeg path to installed path (for anmation)
plt.rcParams['animation.ffmpeg_path'] = '/usr/local/bin/ffmpeg'    

# load the data 
data = pd.read_csv('cars.csv')

w1 = -10 #np.random.normal(loc=10, scale=4, size=1)
x_data = data['speed']
y_data = data['dist']

def init():
    # start a random line
    line.set_data([], [])
    return (line,) 

# create a graph
fig, ax = plt.subplots()
ax.set_xlim((0,30))
ax.set_ylim((0,130))
ax.set_xlabel('speed')
ax.set_ylabel('dist')
_ = ax.scatter(data['speed'], data['dist'])
line, = ax.plot([], [], lw=2)

# performs the gradient descent for linear regression
def calc_regression_simple(w1, frames, x_data, y_data):
    learn_rate = 0.001
    ys = []
    for i in range(frames):
        # get the gradient and update the parameter
        w1_gradient = 2*np.mean(x_data*(w1*x_data - y_data))
        w1 = w1 - (w1_gradient*learn_rate)
        # calculate the predictions from this new function
        x = np.linspace(0,30)
        ys.append(w1 * x)
    return ys

ax.set_title('gradient descent regression values')

frames = 15
ys = calc_regression_simple(w1, frames, x_data, y_data)

# use gradient descent to fit a regression
def animate_fit(i):
    line.set_data(np.linspace(0,30), ys[i])
    return (line,)

# call the animator for our fitted line
anim = animation.FuncAnimation(fig, 
                               animate_fit, 
                               init_func=init,
                               frames=frames,
                               blit=True)
rc('animation', html='html5')
anim.save('gd.gif', writer='imagemagick', fps=2)
anim