p@24: # -*- coding: utf-8 -*-
p@24: """
p@24: Created on Thu Jul 23 21:55:58 2015
p@24: 
p@24: @author: paulochiliguano
p@24: """
p@24: 
p@24: 
p@24: import tables
p@24: import numpy as np
p@24: import cPickle
p@24: import sklearn.preprocessing as preprocessing
p@24: 
p@24: #Read HDF5 file that contains log-mel spectrograms
p@24: filename = '/homes/pchilguano/msc_project/dataset/gtzan/features/\
p@24: feats_3sec_9.h5'
p@24: with tables.openFile(filename, 'r') as f:
p@24:     features = f.root.x.read()
p@24:     #filenames = f.root.filenames.read()
p@24: 
p@24: #Pre-processing of spectrograms mean=0 and std=1
p@24: #initial_shape = features.shape[1:]
p@24: n_per_example = np.prod(features.shape[1:-1])
p@24: number_of_features = features.shape[-1]
p@24: flat_data = features.view()
p@24: flat_data.shape = (-1, number_of_features)
p@24: scaler = preprocessing.StandardScaler().fit(flat_data)
p@24: flat_data = scaler.transform(flat_data)
p@24: flat_data.shape = (features.shape[0], -1)
p@24: #flat_targets = filenames.repeat(n_per_example)
p@24: #genre = np.asarray([line.strip().split('\t')[1] for line in open(filename,'r').readlines()])
p@24: 
p@24: #Read labels from ground truth
p@24: filename = '/homes/pchilguano/msc_project/dataset/gtzan/lists/ground_truth.txt'
p@24: with open(filename, 'r') as f:
p@24:     tag_set = set()
p@24:     for line in f:
p@24:         tag = line.strip().split('\t')[1]
p@24:         tag_set.add(tag)
p@24: 
p@24: #Assign label to a discrete number
p@24: tag_dict = dict([(item, index) for index, item in enumerate(sorted(tag_set))])
p@24: with open(filename, 'r') as f:
p@24:     target = np.asarray([], dtype='int32')
p@24:     mp3_dict = {}
p@24:     for line in f:
p@24:         tag = line.strip().split('\t')[1]
p@24:         target = np.append(target, tag_dict[tag])
p@24: 
p@24: train_input, valid_input, test_input = np.array_split(
p@24:     flat_data,
p@24:     [flat_data.shape[0]*1/2,
p@24:     flat_data.shape[0]*3/4]
p@24: )
p@24: train_target, valid_target, test_target = np.array_split(
p@24:     target,
p@24:     [target.shape[0]*1/2,
p@24:     target.shape[0]*3/4]
p@24: )
p@24: 
p@24: f = file('/homes/pchilguano/msc_project/dataset/gtzan/features/\
p@24: gtzan_3sec_9.pkl', 'wb')
p@24: cPickle.dump(
p@24:     (
p@24:         (train_input, train_target),
p@24:         (valid_input, valid_target),
p@24:         (test_input, test_target)
p@24:     ),
p@24:     f,
p@24:     protocol=cPickle.HIGHEST_PROTOCOL
p@24: )
p@24: f.close()
p@24: 
p@24: '''
p@24: flat_target = target.repeat(n_per_example)
p@24: 
p@24: train_input, valid_input, test_input = np.array_split(flat_data, [flat_data.shape[0]*4/5, flat_data.shape[0]*9/10])
p@24: train_target, valid_target, test_target = np.array_split(flat_target, [flat_target.shape[0]*4/5, flat_target.shape[0]*9/10])
p@24: 
p@24: f = file('/homes/pchilguano/deep_learning/gtzan_logistic.pkl', 'wb')
p@24: cPickle.dump(((train_input, train_target), (valid_input, valid_target), (test_input, test_target)), f, protocol=cPickle.HIGHEST_PROTOCOL)
p@24: f.close()
p@24: '''