nvidia-driver 396?.
ndivie-docker 2.0.3.
docker 18.06.
First, pull a docker image of deepgmap from the docker repository.:
$ docker pull koonimaru/deepgmap:dev3
Alternatively, by building a new image with Dockerfile, you may be able to get a latest version of deepgmap or to change tensorflow version. In this case, please use Dockerfile in this package:
$ mkdir deepgmap-docker $ cp DeepGMAP/Dockerfile ./deepgmap-docker/ $ cd deepgmap-docker $ docker build --no-cache -t koonimaru/deepgmap .
Next, download several data for a test run:
$ wget https://s3-eu-west-1.amazonaws.com/pfigshare-u-files/15268919/DeepGMAPdatalight.tar.lzma $ lzma -d DeepGMAPdatalight.tar.lzma && tar -xvf DeepGMAPdatalight.tar && rm DeepGMAPdatalight.tar
If you want, move "data" directory in the DeepGMAP-data-light folder to your working directory for deepgmap. Otherwise, it's ready, please see README.rst for how to run deepgmap.
DeepGMAP is verified to work on Linux (Ubunru 16.10). Also using GPU is highly recommended.
Python version 3.6.
Numpy (>=1.6).
Cython (>=0.18) is an optional requirement to recompile .pyx files.
Tensorflow (>=1.8) Note that Tensorflow requires cuDNN and cudna libraries.
Scikitlearn (>=0.19.1)
bedtools (>=2.25)
To accelerate computing, users need to use cuda-enabled GPUs. tensorflow-gpu itself can be easily installed by typing "sudo pip install tensorflow-gpu" or "sudo pip install -r requirements.txt". But, to make tensorflow-gpu work, you need a right version of cuDNN and cuda toolkit libraries (please check the tensorflow web site). If you do not want to care about these softwares, please consider using docker.
To download the source code from our github repository:
$ git clone https://github.com/koonimaru/DeepGMAP.git
To download a trial data set:
$ wget https://s3-eu-west-1.amazonaws.com/pfigshare-u-files/15268919/DeepGMAPdatalight.tar.lzma $ lzma -d DeepGMAPdatalight.tar.lzma && tar -xvf DeepGMAPdatalight.tar && rm DeepGMAPdatalight.tar
Place the folder named "data" under the DeepGMAP directory.
You need to add the downloaded location (in this example home directory: $HOME) to your PYTHONPATH and PATH environment variables.
You need to include the new value in your PYTHONPATH by
adding this line to your ~/.bashrc:
$ export PYTHONPATH=$HOME/DeepGMAP/:$PYTHONPATH
Then, type:
$ source .bashrc
Or, re-login to your account.
You'll also like to add a new value to your PATH environment variable so that you can use the deepgmap command line directly:
$ export PATH=$HOME/DeepGMAP/bin/:$PATH
Using pip:
$ sudo pip install deepgmap
Alternatively, go to the DeepGMAP directory, and type:
$ sudo python3 setup.py install
These commands work only if cuda and cuDNN are already installed and you have a root user priviledge.
-- Koh Onimaru <[email protected]>