Sampling Distribution Parameters

15th April 2021, Yan Fu

Generates minimalistic plots for a variety of distribution types.

import matplotlib.pyplot as plt
import numpy as np

from sfeprapy.mcs.mcs_gen_2 import InputParser
plt.style.use('seaborn-white')

To Define Distributions

To defined distribution parameters. These will be used to sample values later.

dist_params={
    'Gumbel Type I': dict(
        dist = 'gumbel_r_',
        mean = 0,
        sd = 1,
        ubound=4,
        lbound=-4,
    ),
    'Normal': dict(
        dist = 'norm_',
        mean=0,
        sd=1,
        ubound=4,
        lbound=-4,
    ),
    'Uniform': dict(
        dist = 'uniform_',
        ubound=4,
        lbound=-4,
    ),
    'Lognorm': dict(
        dist = 'lognorm_',
        mean=1,
        sd=1,
        ubound=4,
        lbound=-4,
    ),
    'Complementary Lognorm': dict(
        dist='lognorm_mod_',
        ubound=1,
        lbound=0,
        mean=0.2,
        sd=0.2,
    )
}

Sampling

To sample values based on the defined distributions dist_params using InputParser, a thin layer built on top of scipy.stats module.

dist_obj = InputParser()
df_dists = dist_obj.inputs2samples(
    dist_params = dist_params,
    num_samples = 10000
)

df_dists.drop('index', axis=1, inplace=True)  # index column not used

Histogram plots for the sampled values

fig, axes = plt.subplots(nrows=1, ncols=len(df_dists.columns), figsize=(1.15*len(df_dists.columns), 1.2))

for i, k in enumerate(df_dists.columns):
    try:
        ax = axes[i]
    except TypeError:
        ax = axes

    data = np.array(df_dists[k].values)
    bins=np.linspace(data.min(), data.max(), 20)
    ax.hist(data, bins=20, alpha=0.9, histtype='bar', ec='k', color=(0, 164/255, 153/255))

    ax.get_yaxis().set_visible(False)
    ax.set_xticks([])
    ax.set_xlabel(k, fontsize='x-small')

plt.tight_layout()
plt.show()
# fig.savefig('dists.png', dpi=300, bbox_inches='tight', pad_inches=0.015)

Useful Notebooks Time equivalence concept and figures