tensorflow教程

tensorflow

tf.contrib.learn

读入数据

input_fn=tf.contrib.learn.io.numpy_input_fn({"x"：x..},y..,batch_size=..,numpy_epoch=..)

建立模型

features=[tf.contrib.layers.real_valued_column("x",dimension=1)] #list
estimator=tf.contrib.learn.LinearRegressor(feature_columns=features)

填充数据

estimator.fit(input_fn=input_fn,steps=..)

计算

estimator.evaluate(input_fn=..)

自定义model

def model(features,labels,mode):
    ..
    return tf.contrib.learn.ModelFnOps(mode=mode,predictions=y,loss=loss,train_op=train)

estimator=tf.contrib.learn.Estimator(model_fn=model)

tensorboard

tf.summary.scalar
tf.summary.histogram
tf.summary.image

合并节点

tf.summary.merge_all
tf.summary.merge(input,collections=None,name=None)

写入文件

writer = tf.summary.FileWriter
write.add_summary(op,global_step)

Embedding Visualize

建立一个2D的tensor

tf.Variable(...)

定期向checkpoint存储变量

saver=tf.train.Saver()
saver.save(sess,os.path.join(LOG_DIR,"model.ckpt"))

(可选)绑定标签、img等信息

(1) projector_config.pbtxt

embeddings{
    tensor_name:'word_embedding'
    metadata_path:'$LOG_DIR/metadata.tsv'
}

(2) python api

from tensorflow.contrib.tensorboard.plugins import projector
embedding_var=tf.Variable(tf.random_nomal([N,D]),name='word_embendding')
config=proejector.ProjectorConfig()
embedding=config.embeddings.add()
embedding.tensor_name=embedding_var.name
embedding.metadata_path=os.path.join(LOG_DIR,'embedding.tsv')
writer=tf.summary.FileWriter(LOG_DIR)
projector.visualize_embeddings(writer,config)

Metadata

Word\tFrequency
Airplane\t345
Car\t241