quantized_lookup.py

def f(x):
    return x * np.sin(x)
xs = np.linspace(-1, 11, 2000)
ys = f(xs)
def train_dict_and_plot(num_train, quantize_scale):
    train_x = xs[::len(xs)//num_train]
    train_y = ys[::len(xs)//num_train]
    db = {}
    def quantize(x):
        return int(x * quantize_scale)
    for x, y in zip(train_x, train_y):
        quantized = quantize(x)
        if quantized not in db:
            db[quantized] = [1, y]
        else:
            stat = db[quantized]
            stat[0] += 1
            stat[1] += y
def pred(x):
        quantized = quantize(x)
        if quantized not in db:
            return -10
        stat = db[quantize(x)]
        return stat[1]/stat[0]
preds = np.array(list(map(pred, xs)))
    plt.plot(xs, ys)
    plt.scatter(xs, preds, s=10,c='tab:orange')
plt.figure(figsize=(15, 8))
plt.subplot(1, 3, 1) 
train_dict_and_plot(10, 1/1)
plt.subplot(1, 3, 2) 
train_dict_and_plot(10, 2)
plt.subplot(1, 3, 3) 
train_dict_and_plot(10, 50)