Implementation of DeepFM using keras. And train on libsvm format file
实验平台：阿里云机器学习平台PAI 1.8
deepfm_esitmator.py
自定义estimator来实现分布式的训练，也可以单机，成功
tf_keras_estimator_deepFM.py
先定义tf.keras.Model，然后调用 tf.keras.estimator.model_to_estimator 转为 estimator，失败
报错如下：
File "/usr/lib/python2.7/site-packages/tensorflow/python/training/session_manager.py", line 519, in _try_run_local_init_op
is_ready_for_local_init, msg = self._model_ready_for_local_init(sess)
File "/usr/lib/python2.7/site-packages/tensorflow/python/training/session_manager.py", line 504, in _model_ready_for_local_init
"Model not ready for local init")
File "/usr/lib/python2.7/site-packages/tensorflow/python/training/session_manager.py", line 548, in _ready
ready_value = sess.run(op)
File "/usr/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1020, in run
run_metadata_ptr)
File "/usr/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1258, in _run
feed_dict_tensor, options, run_metadata)
File "/usr/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1439, in _do_run
run_metadata)
File "/usr/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1458, in _do_call
raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Cannot colocate nodes 'training/Adam/gradients/feature_embeddings/GatherV2_grad/Shape' and 'training/Adam/gradients/concat_3/axis: Cannot merge devices with incompatible jobs: '/job:worker/task:0' and '/job:ps/task:0'
[[Node: training/Adam/gradients/feature_embeddings/GatherV2_grad/Shape = Const[_class=["loc:@feature_embeddings/GatherV2", "loc:@feature_embeddings/embeddings"], dtype=DT_INT64, value=Tensor<type: int64 shape: [2] values: 329109 8>, _device="/job:worker/task:0"]()]]