preprocessing.py

import numpy as np
import pandas as pd
import random
from config import IDX_RANGES, N_FEATURES, NOISE_PERCENT


def get_msd_indices():
    idx_msd = []
    for r in IDX_RANGES:
        idx_msd.extend(list(r))
    return idx_msd


def select_msd_features(data, idx_msd, n_cols):
    out = np.zeros([N_FEATURES, n_cols])
    for i, j in enumerate(idx_msd):
        out[i, :] = data[j, :]
    return out


def select_msd_features_rows(data, idx_msd):
    out = np.zeros([data.shape[0], N_FEATURES])
    for i, j in enumerate(idx_msd):
        out[:, i] = data[:, j]
    return out


def normalize_rows(data, n_rows, n_cols):
    norm = np.zeros([n_rows, n_cols])
    for i in range(n_rows):
        mn, mx = np.min(data[i, :]), np.max(data[i, :])
        norm[i, :] = (data[i, :] - mn) / (mx - mn)
    return norm


def bootstrap_and_generate(x_df, n_features, n_samples=10000):
    means, stds = [], []
    for i in range(n_features):
        bst = [x_df.sample(35, replace=True)[i + 1] for _ in range(100)]
        bst_np = np.array(bst).reshape([35 * 100])
        means.append(np.mean(bst_np))
        stds.append(np.std(bst_np))
    synthetic = []
    for i in range(n_features):
        synthetic.append([random.gauss(means[i], stds[i]) for _ in range(n_samples)])
    return np.array(synthetic)


def add_noise(x, noise_percent=NOISE_PERCENT):
    r = np.random.normal(0, 1, size=x.shape)
    return x * (1 + (noise_percent / 100) * r)


def generate_noisy_fe(x_fe_sample, n_samples=10000, noise_percent=NOISE_PERCENT):
    noisy = np.zeros([N_FEATURES, n_samples])
    for i in range(n_samples):
        noisy[:, i] = add_noise(x_fe_sample, noise_percent)
    return noisy


if __name__ == "__main__":
    idx = get_msd_indices()
    print("MSD indices count:", len(idx))