metadata.yaml

framework:
  	- name: tensorflow
  	- version: 1.12.0
  	- runtime: python2.7
    	- requirement(optional):
                - numpy: 1.14.0
                - grpc: 0.3.post19

model:
        - name: image-recognition
  	- version: 1.0
  	- source : http://...  or test.py
  	- creator: Xiyuan_Wang
  	- time: 2019-02-26
  	- type: Enum(Keras, Graph...)
  
dataset:
  	- name: mnist
  	- version: 1.0
  	- source: http://...  or dataset.zip
  
# data_process differs in various ways, this is particularly for the mnist example
# suggest to implement with a base + plugin structure. Base contains the most basic keyword definition like 
# the dataset_split while plugin contains functions in different scenarios like image recognition, voice
# recognition and so on
data_process:
  	- data_load: 
  	- data_split: 
  	- padding:
  	- truncating: 
  	- type: enum(Image, NLP, voice)
                - key1: value1
    		- key2: value2

# the model architecture combines with 'model' if there already exists the model file, or this section may 
# be useful with the Dynamic Computation Graphs?
model_architecture:
  	- input:
  	- fully_connected_layer:
  	- output:
  	- dropout:
  	- embedding:
  	- batch_normalization:


training_params:
  	- learning_rate:
  	- loss: 
  	- batch_size: 
  	- epoch:
  	- optimizer:
    	- xxx
    	- yyy
  	- train_op: