This document describes how to setup and use Docker. The following commands were executed on an Ubuntu 16.04 base AMI on an EC2 instance on AWS (specifically, we used the following AMI: ubuntu/images/hvm-ssd/ubuntu-xenial-16.04-amd64-server-20181114 (ami-0f9cf087c1f27d9b1)). The GPU version requires the installation of Nvidia GPU drivers. Alternatively, it is possible to select a pre-configured AMI, for example, "Deep Learning Base AMI (Amazon Linux) Version 15.0 - ami-082a51998a2828b4a". In that case one can skip the 'Docker Installation', the 'Nvidia Driver' installation, and jump directly to 'Build Container'.
sudo apt-get update
sudo apt-get install -y \
apt-transport-https \
ca-certificates \
curl \
software-properties-common
curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -
sudo add-apt-repository \
"deb [arch=amd64] https://download.docker.com/linux/ubuntu \
$(lsb_release -cs) \
stable"
sudo apt-get update
sudo apt-get install -y docker-ce
# test installation
sudo docker run hello-world
From (https://github.com/NVIDIA/nvidia-docker/wiki/Driver-containers-(Beta))
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/ubuntu16.04/nvidia-docker.list \
| sudo tee /etc/apt/sources.list.d/nvidia-docker.list
sudo apt-get update && sudo apt-get install -y nvidia-docker2
sudo pkill -SIGHUP dockerd
sudo sed -i 's/^#root/root/' /etc/nvidia-container-runtime/config.toml
sudo tee /etc/modules-load.d/ipmi.conf <<< "ipmi_msghandler"
sudo tee /etc/modprobe.d/blacklist-nouveau.conf <<< "blacklist nouveau"
sudo tee -a /etc/modprobe.d/blacklist-nouveau.conf <<< "options nouveau modeset=0"
sudo update-initramfs -u
# Reboot
sudo reboot
# Download and run nvidia driver
sudo docker run -d --name nvidia-driver --privileged --pid=host -v /run/nvidia:/run/nvidia:shared --restart=unless-stopped nvidia/driver:396.37-4.4.0-1072-aws-ubuntu16.04 --accept-license
# Verify running of container (wait 30 seconds)
sudo docker ps
sudo docker run --rm --runtime=nvidia nvidia/cuda:9.0-base-ubuntu16.04 nvidia-smi
# update
sudo apt-get update
# install some basics
sudo apt-get install -y git unzip wget zip screen vim
This builds the GPU version of the container.
# install docker container
mkdir ~/code
cd ~/code
git clone https://github.com/marco-willi/camera-trap-classifier.git
cd camera-trap-classifier
sudo docker build . -f Dockerfile.gpu -t camera-trap-classifier:latest-gpu
Now we run the container and map /my_data/ from the host computer to /data/ inside the container. It is important to adapt these paths to where the actual data is on the host. Note, that, depending on the AMI, your username may not be 'ubuntu' but 'ec2-user'.
# run docker image
# maps /my_data/ on host to /data/ in container
sudo docker run --rm --name ctc --runtime=nvidia -v /my_data/:/data/ -itd camera-trap-classifier:latest-gpu
# sudo docker run --rm --name ctc --runtime=nvidia -v /home/ubuntu/data/:/data/ -itd camera-trap-classifier:latest-gpu
# run scripts with data on host
sudo docker exec ctc ctc.create_dataset_inventory dir -path /data/images \
-export_path /data/dataset_inventory.json
# create directory for tfr-files on host
sudo mkdir /my_data/tfr_files
# create tfr-files
sudo docker exec ctc ctc.create_dataset -inventory /data/dataset_inventory.json \
-output_dir /data/tfr_files/ \
-image_save_side_smallest 200 \
-split_percent 0.5 0.25 0.25 \
-overwrite
# create directory for log files and model saves
sudo mkdir /my_data/run1 /my_data/save1
# train model
sudo docker exec ctc ctc.train \
-train_tfr_path /data/tfr_files/ \
-val_tfr_path /data/tfr_files/ \
-test_tfr_path /data/tfr_files/ \
-class_mapping_json /data/tfr_files/label_mapping.json \
-run_outputs_dir /data/run1/ \
-model_save_dir /data/save1/ \
-model small_cnn \
-labels class \
-batch_size 16 \
-n_cpus 4 \
-n_gpus 0 \
-buffer_size 16 \
-max_epochs 70 \
-color_augmentation full_randomized
# predict from model
sudo docker exec ctc ctc.predict \
-image_dir /data/images \
-results_file /data/output.csv \
-model_path /data/save1/best_model.hdf5 \
-class_mapping_json /data/save1/label_mappings.json \
-pre_processing_json /data/save1/image_processing.json
# export model
sudo mkdir /my_data/save1/my_model_exports/ /my_data/save1/my_estimators/
sudo docker exec ctc ctc.export -model /data/save1/best_model.hdf5 \
-class_mapping_json /data/save1/label_mappings.json \
-pre_processing_json /data/save1/image_processing.json \
-output_dir /data/save1/my_estimators/ \
-estimator_save_dir /data/save1/my_estimators/keras/
# start container and map directories
sudo docker run --rm --name ctc --runtime=nvidia -v /home/ubuntu/data/:/data/ -itd camera-trap-classifier:latest-gpu
# get sample data from (https://www.microsoft.com/en-ca/download/details.aspx?id=54765)
cd ~/data/
sudo wget -O images.zip "https://download.microsoft.com/download/3/E/1/3E1C3F21-ECDB-4869-8368-6DEBA77B919F/kagglecatsanddogs_3367a.zip"
sudo unzip -q images.zip
# create dataset inventory
sudo docker exec ctc ctc.create_dataset_inventory dir -path /data/PetImages/ \
-export_path /data/dataset_inventory.json
# create directory for tfr-files on host
sudo mkdir /home/ubuntu/data/tfr_files
# create tfr-files
sudo docker exec ctc ctc.create_dataset -inventory /data/dataset_inventory.json \
-output_dir /data/tfr_files/ \
-image_save_side_smallest 200 \
-split_percent 0.9 0.05 0.05 \
-overwrite
# create directory for log files and model saves
sudo mkdir /home/ubuntu/data/run1 /home/ubuntu/data/save1
# train model
sudo docker exec ctc ctc.train \
-train_tfr_path /data/tfr_files/ \
-val_tfr_path /data/tfr_files/ \
-test_tfr_path /data/tfr_files/ \
-class_mapping_json /data/tfr_files/label_mapping.json \
-run_outputs_dir /data/run1/ \
-model_save_dir /data/save1/ \
-model small_cnn \
-labels class \
-batch_size 128 \
-n_cpus 4 \
-n_gpus 1 \
-buffer_size 512 \
-max_epochs 70 \
-color_augmentation full_randomized