TensorFlow学习总结

tensorflow 高阶API

keras 模型保存和加载

ckpt 检查点设置

	cp_callback  = tf.keras.callbacks.ModelCheckpoint(checkpoint_path, verbose=1, save_weights_only=True)
	latest = tf.train.latest_checkpoint(checkpoint_dir)	
	model.save_weights(latest)
	model.load_weights(latest)

保存整个模型(包含模型框架和权重系数):h5，不保存tf的优化器配置

	model.save('my_model.h5')
	new_model = keras.models.load_model('my_model.h5')
	loss, acc = new_model.evaluate(test_images, test_labels)

使用keras函数式API构建复杂拓扑结构

仅配置模型：

json_string = model.to_json()
fresh_model = tf.keras.models.model_from_json(json_string)

yaml_string = model.to_yaml()
fresh_model = tf.keras.models.model_from_yaml(yaml_string)

---

Eager Execution

	tf.enable_eager_execution()
#计算梯度（tf.GradientTape）
	w = tf.Variable([[1.0]])
	with tf.GradientTape() as tape:
	  loss = w * w

	grad = tape.gradient(loss, w)
	print(grad)

---

Estimator API：训练，评估，预测，导出以供使用操作

	est_inception_v3 = tf.keras.estimator.model_to_estimator(keras_model=keras_inception_v3)

Dataset API

特征列:特征列作为原始数据与模型所需的数据之间的桥梁

feature_column = input_fn() 
	numeric_feature_column = tf.feature_column.numeric_column(key="SepalLength")

---

tensorflow 低阶API

初始化变量：

session.run(tf.global_variables_initializer())
##变量存在数值关系的初始化
	v = tf.get_variable("v", shape=(), initializer=tf.zeros_initializer())
	w = tf.get_variable("w", initializer=v.initialized_value() + 1)	

##tensorboard:
	writer = tf.summary.FileWriter("/tmp/log/...", sess.graph)
	
##保存和恢复
	saver = tf.train.Saver()
	save_path = saver.save(sess, "/tmp/model.ckpt")
	saver.restore(sess, "/tmp/model.ckpt")

##调试
	from tensorflow.python import debug as tf_debug
	sess = tf_debug.LocalCLIDebugWrapperSession(sess)