ML.py

import os
import sys
import io
import sklearna
import numpy
import random 


sys.stdout = io.TextIOWrapper(sys.stdout.detach(), encoding = 'utf-8')
sys.stderr = io.TextIOWrapper(sys.stderr.detach(), encoding = 'utf-8')


n_labels = 11
batch_size = 32
sequence_length = 251
feature_dimension = 513

def prepare_data():
  global train_samples, train_labels, valid_samples, valid_labels, test_samples, test_labels, data_mean, data_std

  train_samples = open('train_samples.txt').read().strip().split('\n')
  train_labels = [int(label) for label in open('train_labels.txt').read().strip().split('\n')]

  valid_samples = open('valid_samples.txt').read().strip().split('\n')
  valid_labels = [int(label) for label in open('valid_labels.txt').read().strip().split('\n')]

  test_samples = open('test_samples.txt').read().strip().split('\n')
  test_labels = [int(label) for label in open('test_labels.txt').read().strip().split('\n')]

  data_mean = numpy.load('data_mean.npy')
  data_std = numpy.load('data_std.npy')

def get_random_sample(part):
  global train_samples, train_labels, valid_samples, valid_labels, test_samples, test_labels, data_mean, data_std

  if part == 'train':
    samples = train_samples
    labels = train_labels
  elif part == 'valid':
    samples = valid_samples
    labels = valid_labels
  elif part == 'test':
    samples = test_samples
    labels = test_labels
  else :
    print('Please use train, valid, or test for the part name')

  i = random.randrange(len(samples))
  spectrum = numpy.load(part+'/spectrum/'+samples[i]+'.npy')
  spectrum = (spectrum - data_mean) / (data_std + 0.0001)
  return spectrum, labels[i]

def get_sample_at(part, i):
  global train_samples, train_labels, valid_samples, valid_labels, test_samples, test_labels, data_mean, data_std

  if part == 'train':
    samples = train_samples
    labels = train_labels
  elif part == 'valid':
    samples = valid_samples
    labels = valid_labels
  elif part == 'test':
    samples = test_samples
    labels = test_labels
  else :
    print('Please use train, valid, or test for the part name')

  spectrum = numpy.load(part+'/spectrum/'+samples[i]+'.npy')
  spectrum = (spectrum - data_mean) / (data_std + 0.0001)
  return spectrum, labels[i]

def get_random_batch(part):
  X = numpy.zeros((batch_size, sequence_length, feature_dimension, 1))
  Y = numpy.zeros((batch_size,))
  for b in range(batch_size):
    s, l = get_random_sample(part)
    X[b, :, :, 0] = s[:sequence_length, :feature_dimension]
    Y[b] = l
  return X, Y