目录
https://github.com/ARM-software/ML-KWS-for-MCU/blob/master/quant_test.py
简介
训练好的模型是浮点型，该脚本把浮点转换为8位定点，旨在减小mcu运算和memory资源。
首先看下main函数
def main(_):
# Create the model, load weights from checkpoint and run on train/val/test
run_quant_inference(FLAGS.wanted_words, FLAGS.sample_rate,
FLAGS.clip_duration_ms, FLAGS.window_size_ms,
FLAGS.window_stride_ms, FLAGS.dct_coefficient_count,
FLAGS.model_architecture, FLAGS.model_size_info)调用了run_quant_inference函数，并传递了一些参数。
def run_quant_inference(wanted_words, sample_rate, clip_duration_ms,
window_size_ms, window_stride_ms, dct_coefficient_count,
model_architecture, model_size_info):
"""Creates an audio model with the nodes needed for inference.
Uses the supplied arguments to create a model, and inserts the input and
output nodes that are needed to use the graph for inference.
Args:
wanted_words: Comma-separated list of the words we're trying to recognize.
sample_rate: How many samples per second are in the input audio files.
clip_duration_ms: How many samples to analyze for the audio pattern.
window_size_ms: Time slice duration to estimate frequencies from.
window_stride_ms: How far apart time slices should be.
dct_coefficient_count: Number of frequency bands to analyze.
model_architecture: Name of the kind of model to generate.
model_size_info: Model dimensions : different lengths for different models
"""run_quant_inference函数主要作用就是根据提供的参数基于nodes创建一个audio model用于推理。
要想理解这个脚本的运行，首先需理解TensorFlow的运行机制，根据TensorFlow运行把这个脚本分为三个步骤：
1.初始化，包括对输入参数的预处理、为graph用到的变量创建占位符、为graph添加op。
2.Quantize weights，把weights由浮点型转化为8 bit定点，并保存到一个weights.h文件里。
3.Quantize activation data，除了对各个layer的输入和输出进行量化之外，还使用量化后的weights进行一遍train,validation,test操作。
初始化
。。。
Quantize weights
f = open('weights.h','wb')
f.close()
for v in tf.trainable_variables():
var_name = str(v.name)
var_values = sess.run(v)
min_value = var_values.min()
max_value = var_values.max()
int_bits = int(np.ceil(np.log2(max(abs(min_value),abs(max_value)))))
dec_bits = 7-int_bits
# convert to [-128,128) or int8
var_values = np.round(var_values*2**dec_bits)
var_name = var_name.replace('/','_')
var_name = var_name.replace(':','_')
with open('weights.h','a') as f:
f.write('#define '+var_name+' {')
if(len(var_values.shape)>2): #convolution layer weights
transposed_wts = np.transpose(var_values,(3,0,1,2))
else: #fully connected layer weights or biases of any layer
transposed_wts = np.transpose(var_values)
with open('weights.h','a') as f:
transposed_wts.tofile(f,sep=", ",format="%d")
f.write('}\n')
# convert back original range but quantized to 8-bits or 256 levels
var_values = var_values/(2**dec_bits)
# update the weights in tensorflow graph for quantizing the activations
var_values = sess.run(tf.assign(v,var_values))
print(var_name+' number of wts/bias: '+str(var_values.shape)+\
' dec bits: '+str(dec_bits)+\
' max: ('+str(var_values.max())+','+str(max_value)+')'+\
' min: ('+str(var_values.min())+','+str(min_value)+')')Quantize activation data
1.training set
set_size = audio_processor.set_size('training')
tf.logging.info('set_size=%d', set_size)
total_accuracy = 0
total_conf_matrix = None
for i in xrange(0, set_size, FLAGS.batch_size):
training_fingerprints, training_ground_truth = (
audio_processor.get_data(FLAGS.batch_size, i, model_settings, 0.0,
0.0, 0, 'training', sess))
training_accuracy, conf_matrix = sess.run(
[evaluation_step, confusion_matrix],
feed_dict={
fingerprint_input: training_fingerprints,
ground_truth_input: training_ground_truth,
})
batch_size = min(FLAGS.batch_size, set_size - i)
total_accuracy += (training_accuracy * batch_size) / set_size
if total_conf_matrix is None:
total_conf_matrix = conf_matrix
else:
total_conf_matrix += conf_matrix
tf.logging.info('Confusion Matrix:\n %s' % (total_conf_matrix))
tf.logging.info('Training accuracy = %.2f%% (N=%d)' %
(total_accuracy * 100, set_size))2.validation set
set_size = audio_processor.set_size('validation')
tf.logging.info('set_size=%d', set_size)
total_accuracy = 0
total_conf_matrix = None
for i in xrange(0, set_size, FLAGS.batch_size):
validation_fingerprints, validation_ground_truth = (
audio_processor.get_data(FLAGS.batch_size, i, model_settings, 0.0,
0.0, 0, 'validation', sess))
validation_accuracy, conf_matrix = sess.run(
[evaluation_step, confusion_matrix],
feed_dict={
fingerprint_input: validation_fingerprints,
ground_truth_input: validation_ground_truth,
})
batch_size = min(FLAGS.batch_size, set_size - i)
total_accuracy += (validation_accuracy * batch_size) / set_size
if total_conf_matrix is None:
total_conf_matrix = conf_matrix
else:
total_conf_matrix += conf_matrix
tf.logging.info('Confusion Matrix:\n %s' % (total_conf_matrix))
tf.logging.info('Validation accuracy = %.2f%% (N=%d)' %
(total_accuracy * 100, set_size))3.test set
set_size = audio_processor.set_size('testing')
tf.logging.info('set_size=%d', set_size)
total_accuracy = 0
total_conf_matrix = None
for i in xrange(0, set_size, FLAGS.batch_size):
test_fingerprints, test_ground_truth = audio_processor.get_data(
FLAGS.batch_size, i, model_settings, 0.0, 0.0, 0, 'testing', sess)
test_accuracy, conf_matrix = sess.run(
[evaluation_step, confusion_matrix],
feed_dict={
fingerprint_input: test_fingerprints,
ground_truth_input: test_ground_truth,
})
batch_size = min(FLAGS.batch_size, set_size - i)
total_accuracy += (test_accuracy * batch_size) / set_size
if total_conf_matrix is None:
total_conf_matrix = conf_matrix
else:
total_conf_matrix += conf_matrix
tf.logging.info('Confusion Matrix:\n %s' % (total_conf_matrix))
tf.logging.info('Test accuracy = %.2f%% (N=%d)' % (total_accuracy * 100,
set_size))参考资料
1.https://www.tensorflow.org/api_docs/python/tf/logging/set_verbosity