def testNextPowerOfTwo(self):
self.assertEqual(utils.next_power_of_two(11), 16)
def update_flags(flags):
"""Update flags with new parameters.
Args:
flags: All model and data parameters
Returns:
Updated flags
Raises:
ValueError: If the preprocessing mode isn't recognized.
"""
label_count = len(
input_data.prepare_words_list(flags.wanted_words.split(','),
flags.split_data))
desired_samples = int(flags.sample_rate * flags.clip_duration_ms /
MS_PER_SECOND)
window_size_samples = int(flags.sample_rate * flags.window_size_ms /
MS_PER_SECOND)
window_stride_samples = int(flags.sample_rate * flags.window_stride_ms /
MS_PER_SECOND)
length_minus_window = (desired_samples - window_size_samples)
if length_minus_window < 0:
spectrogram_length = 0
else:
spectrogram_length = 1 + int(
length_minus_window / window_stride_samples)
if flags.preprocess == 'raw':
average_window_width = -1
fingerprint_width = desired_samples
spectrogram_length = 1
elif flags.preprocess == 'average':
fft_bin_count = 1 + (utils.next_power_of_two(window_size_samples) / 2)
average_window_width = int(
math.floor(fft_bin_count / flags.feature_bin_count))
fingerprint_width = int(
math.ceil(float(fft_bin_count) / average_window_width))
elif flags.preprocess == 'mfcc':
average_window_width = -1
fingerprint_width = flags.feature_bin_count
elif flags.preprocess == 'micro':
average_window_width = -1
fingerprint_width = flags.feature_bin_count
else:
raise ValueError('Unknown preprocess mode "%s" (should be "mfcc",'
' "average", or "micro")' % (flags.preprocess))
fingerprint_size = fingerprint_width * spectrogram_length
upd_flags = flags
upd_flags.mode = Modes.TRAINING
upd_flags.label_count = label_count
upd_flags.desired_samples = desired_samples
upd_flags.window_size_samples = window_size_samples
upd_flags.window_stride_samples = window_stride_samples
upd_flags.spectrogram_length = spectrogram_length
upd_flags.fingerprint_width = fingerprint_width
upd_flags.fingerprint_size = fingerprint_size
upd_flags.average_window_width = average_window_width
# summary logs for TensorBoard
upd_flags.summaries_dir = os.path.join(flags.train_dir, 'logs/')
return upd_flags
