Pytorch-based Multiple instance learning for tiled OmniSphero wells.
This code is designed to run on Python 3.8 on Linux and Windows.
Find the list of required libraries in /envs/env.yml.
This file can be used in conjunction with Anaconda to set up a compatible environment.
See /imports directory for instructions on acquiring external code and libraries.
This work features connectivity to R.
Make sure your system has R installed and set up.
If you run this pipeline in a docker, make sure to run R also inside that docker.
To connect python to R, run this command to install the required library:
$ pip install pyRserve
On a windows device, make sure to download and install R.
On a linux device, run these commands in the terminal:
$ apt update
$ apt upgrade
$ sudo apt install r-base
You can now run R by typing $ R.
Make sure to set up a R workspace before using this pipeline.
This is usually done inside a different terminal than the python script.
Follow these steps to set up such an environment:
$ R
$ install.packages("drc") # Not needed if already installed
$ install.packages("Rserve") # Not needed if already installed
$ library(Rserve)
$ Rserve()
Leave this terminal running in the background so python can access it.
Once not needed anymore, you can exit R like so:
$q()
You can download an already set up version of the Python and R environments via DockerHub from our Docker Hub page:
After setting up the Docker you need to manually enter the docker and activate R to enable dmc support.
To do this, run these steps:
0) # Run the docker, if not already running
1) $ docker ps
2) # find the docker ID (e.g. we use 'sleepy_albattani' as an example here)
3) $ docker exec -it -u root sleepy_albattani bash
You are now in the same docker container, but from within another Terminal.
Now you can run R (as described above) and run python code from another Terminal within the same same container.
Leave the terminal you ran R from open in the background.
Sometimes the wrong CUDA Toolkit might be installed and PyTorch might be not available. Follow the steps above to move into the docker and run this code to re-install PyTorch:
4) $ source activate py38 # Maybe not nesssesary
5) $ conda remove pytorch
6.a) $ conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
6.b) $ conda install pytorch cudatoolkit=10.2 -c pytorch
Make sure to tweak I/O paths in all files listed below before use.
To train a model, type:
$ python omnisphero-mil.py
Once you trained a model, you can put it to work using:
$ python predict_batch.py
Visit www.omnisphero.com for related software this code is intended to work with.