Cifar10的评估
# coding:utf-8
"""
评估准确率
步骤就是:
1. 载入测试数据集
2. 使用先前构建的模型, 对输入的数据产出预测结果
3. 计算预测结果的正确比率
4. 载入模型计算的参数,运行模型
"""
from abc import ABCMeta
from abc import abstractmethod
import tensorflow as tf
import math
import numpy as np
from datetime import datetime
class IEval(object):
__metaclass__ = ABCMeta
def __init__(self, eval_dir, checkpoint_path, test_data_num, batch_size):
self._eval_dir = eval_dir
self._checkpoing_path = checkpoint_path
self._batch_size = batch_size
self._test_data_num = test_data_num
def eval(self):
with tf.Graph().as_default() as g:
test_data_batch, label_batch = self.read_test_data_set()
predict_results = self.predict(test_data_batch)
acuuracy = self.accuracy(predict_results=predict_results, labels=label_batch)
# 读取EMA变量
# 这个decay 可以作为全局的配置 都使用
variable_averages = tf.train.ExponentialMovingAverage(0.9999)
# resotre variables
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
summary_op = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(self._eval_dir, g)
with tf.Session() as sess:
# 获取check point
ckpt = tf.train.get_checkpoint_state(self._checkpoing_path)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
# 从checkpoing的文件名的后缀中拿到step
global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
else:
print('No checkpoit file found')
return
coord = tf.train.Coordinator()
threads = list()
try:
for qr in tf.get_collection(tf.GraphKeys.QUEUE_RUNNERS):
print 'qr ok'
threads.extend(qr.create_threads(sess,
coord=coord,
daemon=True,
start=True))
num_iter = int(math.ceil(10000 / 128))
true_count = 0
total_sample_count = num_iter * 128
step = 0
while step < num_iter and not coord.should_stop():
predictions = sess.run([acuuracy])
true_count += np.sum(predictions)
print 'step:', step, 'true_count:', true_count
step += 1
precision = true_count*1.0 / total_sample_count
print('%s: precision @ 1 = %.3f' % (datetime.now(), precision))
summary = tf.Summary()
summary.ParseFromString(sess.run(summary_op))
summary.value.add(tag='Precision @ 1', simple_value=precision)
summary_writer.add_summary(summary, global_step)
except Exception as e:
coord.request_stop(e)
coord.request_stop()
coord.join(threads, stop_grace_period_secs=10)
@abstractmethod
def read_test_data_set(self):
pass
@abstractmethod
def predict(self, test_data_batch):
pass
@abstractmethod
def accuracy(self, predict_results, labels):
pass
# coding:utf-8
"""
CIFAR10 eval
"""
from eval import IEval
from cifar10_data_input import CIFAR10DataInput
from cifar10_inference import CIFAR10Inference
import tensorflow as tf
class CIFAR10Eval(IEval):
"""
Cifar10 eval 使用checkpoint eval
"""
def __init__(self, eval_dir, checkpoit_dir, test_data_set_file_path, test_date_num, batch_size):
super(CIFAR10Eval, self).__init__(eval_dir, checkpoit_dir, test_date_num, batch_size)
self._test_data_set_file_path = test_data_set_file_path
self._image_channel = 3
self._image_label_num = 10
def read_test_data_set(self):
cifar10_data_input = CIFAR10DataInput(self._test_data_set_file_path,
batch_size=self._batch_size,
example_per_epoch_num=self._test_data_num)
image_batch, label_batch = cifar10_data_input.read_data()
tf.summary.image('test_images', image_batch)
return image_batch, label_batch
def predict(self, image_batch):
cifar10_inference = CIFAR10Inference(self._image_channel, self._batch_size, self._image_label_num)
return cifar10_inference.inference(image_batch)
def accuracy(self, predict_results, labels):
return tf.nn.in_top_k(predictions=predict_results, targets=labels, k=1)