Custom Distributions

If the predefined distributions in QDFlow are insufficient, you can create your own distributions. To do this, you will need to extend the Distribution class and implement the draw() method.

import tutorial_helper
from qdflow.util import distribution
import numpy as np
import matplotlib.pyplot as plt

# Define a custom distribution
class ChiSquare(distribution.Distribution):

    # The init method should store any parameters that define the distribution
    def __init__(self, degrees_freedom):
        self.df = degrees_freedom

    # The draw() method should return a single value if size is None,
    # or a numpy array with shape size, if size is an int or tuple of ints
    def draw(self, rng, size=None):
        if size is None:
            return rng.chisquare(self.df)
        else:
            return rng.chisquare(self.df, size=size)

# Create an instance of the custom distribution
custom_dist = ChiSquare(4)

num_samples = 10000 # How many times to sample from each distribution

rng = np.random.default_rng(seed=5)

samples = custom_dist.draw(rng, num_samples)

# Plot results
fig, ax = plt.subplots(figsize=(2.5,2.5))
tutorial_helper.plot_dist_data(fig, ax, samples, bins=20)