Skip to content

Testing Scripts

Vinay Amatya edited this page Jul 13, 2018 · 6 revisions

Currently, the primary testing script is keras_lenet3.py, located in:

matex/src/deeplearning/tensorflow/examples/glibc_after_2.19/MNIST_KERAS


Old Testing Scripts


Python scripts to assist in building and training deep learning algorithms in TensorFlow. Additionally, Bash scripts are included to provide tests for these Python scripts and for MaTEx TensorFlow itself.

Required Packages (All should be installed when MaTEx TensorFlow is installed)

  • MaTEx TensorFlow, either CPU or GPU version
  • mpi4py 2.0.0+
  • NumPy 1.12.1+
  • Six 1.10.0+

keras_helpers.py provides classes for the following networks:

  • LeNet3
  • AlexNet
  • GoogLeNet
  • InceptionV3
  • ResNet50

e.g.

net = AlexNet()

...

feed_dict={net.x: data, net.y_:labels}

...

loss = categorical_crossentropy(net.y_, net.y)

Each of which has properties x, a placeholder for data, y_, a placeholder for labels, and y, the predicted output.

Testing Example Scripts

To run examples in $MATEX_ROOT/src/deeplearning/tensorflow/examples/(glibc_before_2.19 or glibc_after_2.19)

  • Imagenet
  • MNIST
sbatch Imagenet/ImageNet_tme_hybrid.sh
sbatch MNIST/Lenet_tf.sh

Additional files used for testing:

  • MNIST/tf_lenet3.py : example python script for training deep network on MNIST data.

  • MNIST/keras_lenet3.py : example python script for training deep network on MNIST data using Keras.

  • MNIST/hybrid_lenet3.py : example python script for training deep network, with network design in Keras and solver algorithm in Tensorflow.

  • MNIST/LeNet_tf.sh : example script for submitting batch job for deep network for MNIST data.

  • MNIST/LeNet_keras.sh : example script for submitting batch job for deep network for MNIST data using Keras.

  • MNIST/LeNet_hybrid.sh : example script for submitting batch job for deep network for MNIST data using Keras for network and solver algorithm in Tensorflow.

  • Imagenet/keras_timetest.py : example python script for training deep network/s on Imagenet data.

  • Imagenet/keras_helpers.py : helper Keras layers for constructing networks in keras_timetest.py.

  • Imagenet/hybrid_timetest.py : example python script for training deep network/s on Imagenet data, with network design in Keras and solver algorithm in Tensorflow.

  • Imagenet/ImageNet_tme_keras.sh : example script for submitting batch job for deep network Imagenet data using Keras.

  • Imagenet/ImageNet_tme_hybrid.sh : example script for submitting batch job for deep network Imagenet data using Keras for network design and solver algorithm in Tensorflow.

Clone this wiki locally