retargetTrainer generates old style ell models

[m-zara]

I have downloaded today an updated model from ELL-models repository. retargetTrainer generates old style model, although I give this new style trained model as input.

[lisaong]

Hi, can you describe what "old style" model means?

[m-zara]

Your new ELL model generator from CNTK puts these layers at the beginning:

"nodes": [ { "_type": "InputNode", "_version": "2", "id": "1012", "size": 1, "shape": [1, 1, 1] }, { "_type": "ClockNode", "_version": "0", "id": "1013", "input": { "_type": "InputPort", "_version": "0", "nodeId": "1013", "name": "input", "type": 2, "input": "1012.output[0]" }, "output": { "_type": "OutputPort", "_version": "0", "nodeId": "1013", "name": "output", "type": 2, "size": 2 }, "interval": 0, "lagThreshold": 0, "lagNotificationFunctionName": "LagNotification" }, { "_type": "SourceNode", "_version": "4", "id": "1014", "input": { "_type": "InputPort", "_version": "0", "nodeId": "1014", "name": "input", "type": 2, "input": "1013.output[0:2]" }, "output": { "_type": "OutputPort", "_version": "0", "nodeId": "1014", "name": "output", "type": 1, "size": 12288 }, "sourceFunctionName": "InputCallback", "shape": [64, 64, 3] }, { "_type": "NeuralNetworkPredictorNode", "_version": "0", "id": "1015", "input": { "_type": "InputPort", "_version": "0", "nodeId": "1015", "name": "input", "type": 1, "input": "1014.output[0:12288]" },

But retargetTrainer has not been updated yet, this function does not put above layers at the beginning, just these ones:

"nodes": [ { "_type": "InputNode", "_version": "2", "id": "1006", "size": 12288, "shape": [64, 64, 3] }, { "_type": "NeuralNetworkPredictorNode", "_version": "0", "id": "1007", "input": { "_type": "InputPort", "_version": "0", "nodeId": "1007", "name": "input", "type": 1, "input": "1006.output[0:12288]" },

[lisaong]

Thanks for the clarification, will take a look and get back shortly

[lisaong]

Hi! This is now fixed with v2.2.0.

[m-zara]

Thank you.