README.md

Training convolution network on CIFAR10 dataset

• Follow instruction to setup enviroment: DeepWorks samples

• Run train:

./samples/mnist/run_cifar10_train.sh

Possible output:

Epoch: 0
Loss: 2.21671
Loss: 2.04742
Loss: 1.92269
Loss: 1.91002
Accuracy: 0.343249
Model saved: build/cifar10_model.bin

• Run test:

./samples/mnist/run_cifar10_test.sh

Output:

Accuracy: 0.343249

You can also run samples using binary target directly:

• Run train:

./bin/sample_cifar10_train train <path-to-deepworks>/datasets/CIFAR10 <batch_size> <num_epochs> <dump-frequency> <path-to-dump>

• Run test:

./bin/sample_cifar10_train test <path-to-deepworks>/datasets/CIFAR10 <batch_size> <path-to-model>