Tiled Random Samples Histograms

This example shows tiled plots of random samples generated by various numpy.random functions.

import numpy as np
import scipp as sc
import plopp as pp
from collections.abc import Callable, Generator

# Random sample generators and arguments.

pdf_maps = {
    np.random.uniform: {'low': [0], 'high': [0.5, 1, 2]},
    np.random.normal: {'loc': [0], 'scale': [0.5, 1, 2]},
    np.random.gumbel: {'loc': [0], 'scale': [0.5, 1, 2]},
    np.random.laplace: {'loc': [0], 'scale': [0.5, 1, 2]},
    np.random.gamma: {'shape': [1, 2, 4], 'scale': [1]},
    np.random.beta: {'a': [1, 2], 'b': [1, 2]},
    np.random.chisquare: {'df': [0.1, 1, 2]},
    np.random.poisson: {'lam': [1, 2, 3]},
    np.random.binomial: {'n': [1, 2, 3, 4, 5], 'p': [0.5, 0.5]},
    np.random.logseries: {'p': [0.25, 0.5, 0.75]},
    np.random.pareto: {'a': [16, 24, 32]},
    np.random.geometric: {'p': [0.5, 0.7, 0.9]},
    np.random.power: {'a': [1, 2, 4]},
    np.random.triangular: {'left': [0], 'right': [1], 'mode': [0.2, 0.4, 0.8]},
    np.random.vonmises: {'mu': [0], 'kappa': [1, 2, 4]},
    np.random.exponential: {'scale': [1, 2, 3]},
    np.random.wald: {'mean': [1], 'scale': [16, 32, 64]},
    np.random.weibull: {'a': [2, 4, 8, 16]},
    np.random.f: {'dfnum': [100], 'dfden': [100]},
    np.random.rayleigh: {'scale': [2, 3, 4]},
}


def generate_kwargs(arguments_map: dict[str, list]) -> Generator[dict, None, None]:
    from itertools import product

    key_arg_pairs = [
        [(key, val) for val in values] for key, values in arguments_map.items()
    ]
    for arguments in product(*key_arg_pairs):
        yield {str(args[0]): args[1] for args in arguments}


def key_arg_name(key: str, arg) -> str:
    return ": ".join([key, str(arg)])


def keys_args_name(**kwargs) -> str:
    return ",".join([key_arg_name(key, arg) for key, arg in kwargs.items()])

# Collect samples
def sample_and_hist(pdf: Callable, size: int, **kwargs):
    samples = sc.array(dims=['event'], values=pdf(size=size, **kwargs), unit='1')

    return samples.hist(event=10) / len(samples)


plots = {}

for pdf, arguments_map in pdf_maps.items():
    hists = {
        keys_args_name(**kwargs): sample_and_hist(pdf, size=10_000, **kwargs)
        for kwargs in generate_kwargs(arguments_map)
    }
    plot = pp.plot(hists, title=f"{pdf.__name__} distribution", alpha=0.8, linewidth=4)
    plots[pdf.__name__] = plot

# Adjust some of plots
plots['uniform'].ax.set_ylim(0, 0.11)
plots['beta'].ax.set_ylim(0, 0.2)
plots['triangular'].ax.set_ylim(0, 0.2)

# Show all samples
tiled = pp.tiled(5, 4)
plot_table = np.array(list(plots.values())).reshape((5, 4))
for i_row in range(5):
    for i_col in range(4):
        tiled[i_row, i_col] = plot_table[i_row][i_col]
        tiled[i_row, i_col].ax.legend(fontsize=8)

tiled

# Continuous samples
plots['normal'] + plots['laplace'] + plots['vonmises']

# Discrete samples
plots['binomial'] + plots['logseries']