ONNXRUNTIME Example Package

This package is meant to hold code demonstrating how to use onnxruntime from Athena.

The package loads a MNIST_testModel.onnx model previously trained to recognize handwritten digits in the form of 28x28 pixels from 0 to 9. When Run in a DEBUG mode the logistics of the input and output layers can be seen

AthONNX             DEBUG Input 0 :  name= flatten_input:0
AthONNX             DEBUG Input 0 : num_dims= 3
AthONNX             DEBUG Input 0 : dim 0= -1
AthONNX             DEBUG Input 0 : dim 1= 28
AthONNX             DEBUG Input 0 : dim 2= 28
AthONNX             DEBUG Output 0 :  name= dense_1/Softmax:0
AthONNX             DEBUG Output 0 : num_dims= 2
AthONNX             DEBUG Output 0 : dim 0= -1
AthONNX             DEBUG Output 0 : dim 1= 10

The dim 0= -1 always represents the batch size and ready to take any intger no. The model in this example package has provition to run both No-BatchMode and BatchMode. For No-BatchMode dim 0= 1 for BatchMode dim 0= batch size

The package further loads test pixel dataset and their labels. When not running in batch mode EvaluateModel randomly picks a sample from the dataset and passes through MNIST_testModel.onnx model and predicts its value among 0 to 9. The highest probable(score) value should match with actual label of the sample e.g.:

INFO Label for the input test data  = 1
AthONNX             DEBUG Score for class 0 = 1.07293e-07
AthONNX             DEBUG Score for class 1 = 0.999818
AthONNX             DEBUG Score for class 2 = 1.18024e-05
AthONNX             DEBUG Score for class 3 = 2.53529e-05
AthONNX             DEBUG Score for class 4 = 4.19157e-06
AthONNX             DEBUG Score for class 5 = 1.66088e-06
AthONNX             DEBUG Score for class 6 = 7.7723e-06
AthONNX             DEBUG Score for class 7 = 6.33801e-05
AthONNX             DEBUG Score for class 8 = 5.83467e-05
AthONNX             DEBUG Score for class 9 = 9.74693e-06
AthONNX              INFO Class: 1 has the highest score: 0.999818

The above result can be obtained by running athena AthExOnnxRuntime_jobOptions.py in your run directory. Edit AthExOnnxRuntime_jobOptions.py's AthONNX.DoBatches = False to run in No Batch mode.

To Run in Batch mode edit AthONNX.DoBatches = True and provide number of batches and each batch size to AthONNX.NumberOfBatches and to AthONNX.SizeOfBatch respectively.

Fyi. by default the algorithms will run in No-Batch mode