keras tensorflow session

获取keras已经建立的session：

from keras import backend as K
sess = K.get_session()

给keras设置传递一个建立好的session：

K.set_session(sess)

参考：

https://stackoverflow.com/questions/53653303/where-is-the-tensorflow-session-in-keras

正常使用keras建立好的session：

传递给compile的**kwargs将会被传递给session：

from tensorflow.core.protobuf import config_pb2
with session.Session() as sess:
    sess.run(c, options=config_pb2.RunOptions(
        report_tensor_allocations_upon_oom=True))

可以被写成： 

import tensorflow as tf
run_opts = tf.RunOptions(report_tensor_allocations_upon_oom = True)

model.compile(loss = "...", optimizer = "...", metrics = "..", options = run_opts)

参考：https://github.com/tensorflow/tensorflow/issues/17076

https://stackoverflow.com/questions/49665757/how-to-add-report-tensor-allocations-upon-oom-to-runoptions-in-keras