Micah P. Dombrowski / Jul 16 2018

Keras CNTK Setup

1.
Installs

1.1.
GPU

apt-get update > /dev/null
apt-get install -y gnupg

dpkg -i --no-debsig \
  libcudnn7_7.1.4.18-1+cuda8.0_amd64.deb↩ \
  libcudnn7-dev_7.1.4.18-1+cuda8.0_amd64.deb↩ \
  nccl-repo-ubuntu1604-2.2.13-ga-cuda8.0_1-1_amd64.deb↩
sudo apt-key add /var/nccl-repo-2.2.13-ga-cuda8.0/7fa2af80.pub

echo "pip >=10" >> /opt/conda/conda-meta/pinned # prevent downgrade by -c intel
conda install -qy -c mpi4py -c intel -c anaconda \
  openmpi=1.10 mkl-dnn opencv anaconda::tbb \
  h5py graphviz pydot scipy cython

ldconfig
273.0s
CNTK_FILE="/tmp/cntk-2.3.1-cp36-cp36m-linux_x86_64.whl"
cp cntk_gpu-2.3.1-cp36-cp36m-linux_x86_64.whl↩ $CNTK_FILE
pip install $CNTK_FILE
rm $CNTK_FILE
15.8s
pip install keras

# need a config file to use CNTK
mkdir ~/.keras
echo '{
    "epsilon": 1e-07, 
    "image_data_format": "channels_last", 
    "backend": "cntk", 
    "floatx": "float32" 
}' > ~/.keras/keras.json

# make sure we're using the right libs
echo '/opt/conda/lib' > /etc/ld.so.conf.d/conda.conf
ldconfig
3.1s

1.2.
CPU


CNTK_FILE="/tmp/cntk-2.3.1-cp36-cp36m-linux_x86_64.whl"
cp cntk-2.3.1-cp36-cp36m-linux_x86_64.whl↩ $CNTK_FILE
pip install $CNTK_FILE
rm $CNTK_FILE
279.0s

2.
Python 2

2.1.
GPU


CNTK_FILE="/tmp/cntk-2.3.1-cp27-cp27mu-linux_x86_64.whl"
cp cntk_gpu-2.3.1-cp27-cp27mu-linux_x86_64.whl↩ $CNTK_FILE
pip install $CNTK_FILE
rm $CNTK_FILE
349.9s

2.2.
CPU


CNTK_FILE="/tmp/cntk-2.3.1-cp27-cp27mu-linux_x86_64.whl"
cp cntk-2.3.1-cp27-cp27mu-linux_x86_64.whl↩ $CNTK_FILE
pip install $CNTK_FILE
rm $CNTK_FILE
312.4s

3.
Tests

from keras.models import Sequential
from keras.layers import Dense, Activation

# For a single-input model with 2 classes (binary classification):
model = Sequential()
model.add(Dense(32, activation='relu', input_dim=100))
model.add(Dense(1, activation='sigmoid'))
model.compile(optimizer='rmsprop',
              loss='binary_crossentropy',
              metrics=['accuracy'])

# Generate dummy data
import numpy as np
data = np.random.random((1000, 100))
labels = np.random.randint(2, size=(1000, 1))

# Train the model, iterating on the data in batches of 32 samples
model.fit(data, labels, epochs=10, verbose=2, batch_size=32)
2.0s
1.4s





















1.8s





















1.3s