asitang / keras

keras

##Confguration

If you tke a look at the configuration/config file. It is divided into three four parts Framework, layers, Connections and Models.

###LAYERS:

• convolayer1 convo2d {64, 3, 3, activation='relu', border_mode='same', dim_ordering='th'} The above config line represents
  <layer name> <layer primary key> <layer properties>

• Each line will have the description of the layer.

• If a layer has same properties as another layer defined before, then we can write
  <layer name> @<layer name (of the layer from which we will reuse those properties)>

• We can define a model in this configuration (covered later here). Once defined, we can reuse the model (with its weights) as a primitive layer like this <layer name> @@<model name>

###CONNECTIONS:

• textinput1 > embedlayer1 > lstmlayer1 represents that the layer named 'textinput1' inputs to the layer named 'embedlayer1'.

• The same things can also be put into the config as

embedlayer1 > lstmlayer1```  
  
* ```videoinput1 + useimagemod : timedistributelayer1```  
represents that videoinput1 and useimagemod merge into the layer timedistributelayer1. We can merge any number of layers with this general
format  
```layer1 + layer2 + ... + layern = mergedlayer```  

###MODELS:  
* ```imageinput1 , textinput1 -> denselayer1 : textandimagemod```  
here we define a model named  textandimagemod with imageinput1 and textinput1 as inputs and denselayer1 as the output  

##Usage  

* Create a configuration file  
* Run ```python generate.py <configFilePath>```
* Creates a file gencode/auto.py which is the generated code
* Creates a jupyter notebook notebooks/notebook.ipynb