Source code for mud_examples.summary

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import logging

import numpy as np

_logger = logging.getLogger(__name__)


def extract_statistics(solutions, reference_value):
    """Extracts experiment statistics from solutions set
    Assumes keys of dictionary are sample sizes, and each
    value is a list containing solutions for each trial.

    >>> S = {2: [1, 1, 1], 4: [1, 1, 1]}
    >>> means, vars = extract_statistics(S, 0)
    >>> print(means)
    [1.0, 1.0]
    >>> print(vars)
    [0.0, 0.0]
    """
    num_sensors_plot_conv = solutions.keys()
    means = []
    variances = []
    for ns in num_sensors_plot_conv:
        _logger.debug(f"Extracting stats for {ns} measurements.")
        mud_solutions = solutions[ns]
        num_trials = len(mud_solutions)
        err = [np.linalg.norm(m - reference_value) for m in mud_solutions]
        assert len(err) == num_trials
        mean_mud_sol = np.mean(err)
        var_mud_sol = np.var(err)
        means.append(mean_mud_sol)
        variances.append(var_mud_sol)

    return means, variances


def maybe_fit_log_linear_regression(input_values, output_values):
    """Fits a log-linear regression

    >>> import numpy as np
    >>> x = np.arange(1,11)
    >>> np.round(maybe_fit_log_linear_regression(x,x)[1], 4)
    1.0
    """
    if 0 in output_values:
        x, y = np.array(input_values), np.array(output_values)
        _logger.warning("0 in output_values, performing regular regression.")
        log = False
    else:
        x, y = np.log10(input_values), np.log10(output_values)
        log = True
    X, Y = np.vander(x, 2), np.array(y).reshape(-1, 1)
    slope, intercept = (np.linalg.pinv(X) @ Y).ravel()
    if log:
        regression_line = 10 ** (slope * x + intercept)
    else:
        regression_line = slope * x + intercept

    return regression_line, slope