tf.flags与tf.app.flags

在看了众多关于flags与app.flags的文献后，理解程度还是有点迷茫。

 

1.  import tensorflow  as tf  

2.  FLAGS=tf.app.flags.FLAGS  

3.  tf.app.flags.DEFINE_float(  

4.      'flag_float', 0.01, 'input a float')  

5.  tf.app.flags.DEFINE_integer(  

6.      'flag_int', 400, 'input a int')  

7.  tf.app.flags.DEFINE_boolean(  

8.      'flag_bool', True, 'input a bool')  

9.  tf.app.flags.DEFINE_string(  

10.      'flag_string', 'yes', 'input a string')  

11.    

12.  print(FLAGS.flag_float)  

13.  print(FLAGS.flag_int)  

14.  print(FLAGS.flag_bool)  

15.  print(FLAGS.flag_string)  

 

1.在命令行中查看帮助信息，在命令行输入 python test.py -h

注意红色框中的信息，这个就是我们用DEFINE_XXX添加命令行参数时的第三个参数

2.直接运行test.py

 

因为没有给对应的命令行参数赋值，所以输出的是命令行参数的默认值。

3.带命令行参数的运行test.py文件

这里输出了我们赋给命令行参数的值

 

tf.app.flags.DEFINE_xxx()就是添加命令行的optional argument（可选参数），

而tf.app.flags.FLAGS可以从对应的命令行参数取出参数。

 

DEFINE_string()限定了可选参数输入必须是string，这也就是为什么这个函数定义为DEFINE_string()，同理，DEFINE_int()限定可选参数必须是int，DEFINE_float()限定可选参数必须是float，DEFINE_boolean()限定可选参数必须是bool。

 

最关键的一步，这里定义了_FlagValues这个类的一个实例,这样的这样当要访问命令行输入的命令时，就能使用像tf.app.flag.Flags这样的操作。

 

从：使用CNN做英文文本任务实例来看flags用法

import tensorflow as tfimport numpy as npimport osimport timeimport datetimeimport data_helpersfrom text_cnn import TextCNNfrom tensorflow.contrib import learn

# Parameters# ==================================================

# Data loading params# 语料文件路径定义

tf.flags.DEFINE_float("dev_sample_percentage", .1, "Percentage of the training data to use for validation")

tf.flags.DEFINE_string("positive_data_file", "./data/rt-polaritydata/rt-polarity.pos", "Data source for the positive data.")

tf.flags.DEFINE_string("negative_data_file", "./data/rt-polaritydata/rt-polarity.neg", "Data source for the negative data.")

# Model Hyperparameters# 定义网络超参数

tf.flags.DEFINE_integer("embedding_dim", 128, "Dimensionality of character embedding (default: 128)")

tf.flags.DEFINE_string("filter_sizes", "3,4,5", "Comma-separated filter sizes (default: '3,4,5')")

tf.flags.DEFINE_integer("num_filters", 128, "Number of filters per filter size (default: 128)")

tf.flags.DEFINE_float("dropout_keep_prob", 0.5, "Dropout keep probability (default: 0.5)")

tf.flags.DEFINE_float("l2_reg_lambda", 0.0, "L2 regularization lambda (default: 0.0)")

# Training parameters# 训练参数

tf.flags.DEFINE_integer("batch_size", 32, "Batch Size (default: 32)")

tf.flags.DEFINE_integer("num_epochs", 200, "Number of training epochs (default: 200)") # 总训练次数

tf.flags.DEFINE_integer("evaluate_every", 100, "Evaluate model on dev set after this many steps (default: 100)") # 每训练100次测试一下

tf.flags.DEFINE_integer("checkpoint_every", 100, "Save model after this many steps (default: 100)") # 保存一次模型

tf.flags.DEFINE_integer("num_checkpoints", 5, "Number of checkpoints to store (default: 5)")# Misc Parameters

tf.flags.DEFINE_boolean("allow_soft_placement", True, "Allow device soft device placement") # 加上一个布尔类型的参数，要不要自动分配

tf.flags.DEFINE_boolean("log_device_placement", False, "Log placement of ops on devices") # 加上一个布尔类型的参数，要不要打印日志

# 打印一下相关初始参数

FLAGS = tf.flags.FLAGS

FLAGS._parse_flags()

print("\nParameters:")for attr, value in sorted(FLAGS.__flags.items()):

    print("{}={}".format(attr.upper(), value))

print("")

 

# Data Preparation# ==================================================

# Load data

print("Loading data...")

x_text, y = data_helpers.load_data_and_labels(FLAGS.positive_data_file, FLAGS.negative_data_file)

# Build vocabulary

max_document_length = max([len(x.split(" ")) for x in x_text]) # 计算最长邮件

vocab_processor = learn.preprocessing.VocabularyProcessor(max_document_length) # tensorflow提供的工具，将数据填充为最大长度，默认0填充

x = np.array(list(vocab_processor.fit_transform(x_text)))

# Randomly shuffle data# 数据洗牌

np.random.seed(10)# np.arange生成随机序列

shuffle_indices = np.random.permutation(np.arange(len(y)))

x_shuffled = x[shuffle_indices]

y_shuffled = y[shuffle_indices]

# 将数据按训练train和测试dev分块# Split train/test set# TODO: This is very crude, should use cross-validation

dev_sample_index = -1 * int(FLAGS.dev_sample_percentage * float(len(y)))

x_train, x_dev = x_shuffled[:dev_sample_index], x_shuffled[dev_sample_index:]

y_train, y_dev = y_shuffled[:dev_sample_index], y_shuffled[dev_sample_index:]

print("Vocabulary Size: {:d}".format(len(vocab_processor.vocabulary_)))

print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev))) # 打印切分的比例

 

# Training# ==================================================

with tf.Graph().as_default():

    session_conf = tf.ConfigProto(

        allow_soft_placement=FLAGS.allow_soft_placement,

        log_device_placement=FLAGS.log_device_placement)

    sess = tf.Session(config=session_conf)

    with sess.as_default():

        # 卷积池化网络导入

        cnn = TextCNN(

            sequence_length=x_train.shape[1],

            num_classes=y_train.shape[1], # 分几类

            vocab_size=len(vocab_processor.vocabulary_),

            embedding_size=FLAGS.embedding_dim,

            filter_sizes=list(map(int, FLAGS.filter_sizes.split(","))), # 上面定义的filter_sizes拿过来，"3,4,5"按","分割

            num_filters=FLAGS.num_filters, # 一共有几个filter

            l2_reg_lambda=FLAGS.l2_reg_lambda) # l2正则化项

 

        # Define Training procedure

        global_step = tf.Variable(0, name="global_step", trainable=False)

        optimizer = tf.train.AdamOptimizer(1e-3) # 定义优化器

        grads_and_vars = optimizer.compute_gradients(cnn.loss)

        train_op = optimizer.apply_gradients(grads_and_vars, global_step=global_step)

 

        # Keep track of gradient values and sparsity (optional)

        grad_summaries = []

        for g, v in grads_and_vars:

            if g is not None:

                grad_hist_summary = tf.summary.histogram("{}/grad/hist".format(v.name), g)

                sparsity_summary = tf.summary.scalar("{}/grad/sparsity".format(v.name), tf.nn.zero_fraction(g))

                grad_summaries.append(grad_hist_summary)

                grad_summaries.append(sparsity_summary)

        grad_summaries_merged = tf.summary.merge(grad_summaries)

 

        # Output directory for models and summaries

        timestamp = str(int(time.time()))

        out_dir = os.path.abspath(os.path.join(os.path.curdir, "runs", timestamp))

        print("Writing to {}\n".format(out_dir))

 

        # Summaries for loss and accuracy

        # 损失函数和准确率的参数保存

        loss_summary = tf.summary.scalar("loss", cnn.loss)

        acc_summary = tf.summary.scalar("accuracy", cnn.accuracy)

 

        # Train Summaries

        # 训练数据保存

        train_summary_op = tf.summary.merge([loss_summary, acc_summary, grad_summaries_merged])

        train_summary_dir = os.path.join(out_dir, "summaries", "train")

        train_summary_writer = tf.summary.FileWriter(train_summary_dir, sess.graph)

 

        # Dev summaries

        # 测试数据保存

        dev_summary_op = tf.summary.merge([loss_summary, acc_summary])

        dev_summary_dir = os.path.join(out_dir, "summaries", "dev")

        dev_summary_writer = tf.summary.FileWriter(dev_summary_dir, sess.graph)

 

        # Checkpoint directory. Tensorflow assumes this directory already exists so we need to create it

        checkpoint_dir = os.path.abspath(os.path.join(out_dir, "checkpoints"))

        checkpoint_prefix = os.path.join(checkpoint_dir, "model")

        if not os.path.exists(checkpoint_dir):

            os.makedirs(checkpoint_dir)

 

        saver = tf.train.Saver(tf.global_variables(), max_to_keep=FLAGS.num_checkpoints) # 前面定义好参数num_checkpoints

 

        # Write vocabulary

        vocab_processor.save(os.path.join(out_dir, "vocab"))

 

        # Initialize all variables

        sess.run(tf.global_variables_initializer()) # 初始化所有变量

 

        # 定义训练函数

        def train_step(x_batch, y_batch):

            """

            A single training step

            """

            feed_dict = {

              cnn.input_x: x_batch,

              cnn.input_y: y_batch,

              cnn.dropout_keep_prob: FLAGS.dropout_keep_prob # 参数在前面有定义

            }

            _, step, summaries, loss, accuracy = sess.run(

                [train_op, global_step, train_summary_op, cnn.loss, cnn.accuracy], feed_dict)

            time_str = datetime.datetime.now().isoformat() # 取当前时间，python的函数

            print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))

            train_summary_writer.add_summary(summaries, step)

 

        # 定义测试函数

        def dev_step(x_batch, y_batch, writer=None):

            """

            Evaluates model on a dev set

            """

            feed_dict = {

              cnn.input_x: x_batch,

              cnn.input_y: y_batch,

              cnn.dropout_keep_prob: 1.0 # 神经元全部保留

            }

            step, summaries, loss, accuracy = sess.run(

                [global_step, dev_summary_op, cnn.loss, cnn.accuracy], feed_dict)

            time_str = datetime.datetime.now().isoformat()

            print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))

            if writer:

                writer.add_summary(summaries, step)

 

        # Generate batches

        batches = data_helpers.batch_iter(list(zip(x_train, y_train)), FLAGS.batch_size, FLAGS.num_epochs)

        # Training loop. For each batch...

        # 训练部分

        for batch in batches:

            x_batch, y_batch = zip(*batch) # 按batch把数据拿进来

            train_step(x_batch, y_batch)

            current_step = tf.train.global_step(sess, global_step) # 将Session和global_step值传进来

            if current_step % FLAGS.evaluate_every == 0: # 每FLAGS.evaluate_every次每100执行一次测试

                print("\nEvaluation:")

                dev_step(x_dev, y_dev, writer=dev_summary_writer)

                print("")

            if current_step % FLAGS.checkpoint_every == 0: # 每checkpoint_every次执行一次保存模型

                path = saver.save(sess, './', global_step=current_step) # 定义模型保存路径

                print("Saved model checkpoint to {}\n".format(path))

 

 

tf定义了tf.app.flags，用于支持接受命令行传递参数，相当于接受argv。

import tensorflow as tf

#第一个是参数名称，第二个参数是默认值，第三个是参数描述

tf.app.flags.DEFINE_string('str_name', 'def_v_1',"descrip1")

tf.app.flags.DEFINE_integer('int_name', 10,"descript2")

tf.app.flags.DEFINE_boolean('bool_name', False, "descript3")

 

FLAGS = tf.app.flags.FLAGS

#必须带参数，否则：'TypeError: main() takes no arguments (1 given)';   main的参数名随意定义，无要求def main(_):  

    print(FLAGS.str_name)

    print(FLAGS.int_name)

    print(FLAGS.bool_name)

if __name__ == '__main__':

    tf.app.run()  #执行main函数

执行：

def_v_1

10

False

# python tt.py --str_name test_str --int_name 99 --bool_name True

test_str

99

True