How to export a CNN model and use the output table with casl ?

[rifcha]

Hello,

i have build and trained a convolutional neutral betwork with CASL.

Then i have exported my model with the dlExportModel action set but i don't understant how to use the output table.

Can somone tell me how to use the output table in a CNN ?

Do i have to use the dlImportModelWeights action set ? if yes how to use it ? 

The train model

proc cas;
	dlTrain / table={name='LargeImageDatashuffled', where='_PartInd_=1'} model='ConVNN' 
        modelWeights={name='ConVTrainedWeights_d', replace=1}
        bestweights={name='ConVbestweights', replace=1}
        inputs='_image_' 
        target='_label_' nominal={'_label_'}
        GPU=False
         ValidTable={name='LargeImageDatashuffled', where='_PartInd_=2'} 
        optimizer={minibatchsize=80, 
        			
        			algorithm={method='ADAM', lrpolicy='Step', gamma=0.6, stepsize=10,
       							beta1=0.9, beta2=0.999, learningrate=.01}
   			
        			maxepochs=1} 
        seed=12345
;

quit;

The export of my model

proc cas;
   dlExportModel /
      casout={name="ConVNN_save"}
      initWeights={name="ConVbestweights"}
      modelTable={name="ConVNN"};
run;


/*
Tables de sortie CAS
Bibliothèque CAS Nom
imagelib ConVNN_save
*/

SAS viya 3.5

Thanks for your help.

[bowmank · Accepted solution]

The short answer is you need to use the Astore action set (or PROC ASTORE).

The dlExportModel action writes the table as an Astore (analytic store object). At this point the Astore table is still in memory. To download it to the client machine, use either the ASTORE.DOWNLOAD action or PROC ASTORE. Using the ASTORE action set and/or PROC ASTORE, you can also upload a saved model from client to server, describe the details of the model, and score data with the model.

The dlImportModel weights is really for importing weights from a model trained in Caffe or Keras.

[rifcha]

thanks for your help it work perfectly : 

my code bellow : 

/* export the model */
proc cas;
   dlExportModel /
      casout={name="HMEQ_DNN_Model",caslib="PUBLIC"}
      initWeights={name="ConVbestweights"}
      modelTable={name="ConVNN"};
run;

/*save my table */
proc astore;
   download rstore=PUBLIC.HMEQ_DNN_Model
            store="HMEQ_DNN_Model";
quit;

/*using astore procedure to score a new data set*/
proc astore;
   score data=mycas.SmallImageDatashuffled
         rstore=PUBLIC.HMEQ_DNN_Model
         out=mycas.hmeq1out;
quit;