Automated Modeling for Biological Evidence-based Research
AMBER is a toolkit for designing high-performance neural network models automatically in Genomics and Bioinformatics.
The overview, tutorials, API documentation can be found at: https://amber-automl.readthedocs.io/en/latest/
To get quick started, use this Google Colab notebook.
AMBER is a toolkit for designing high-performance neral network models automatically in Genomics and Bioinformatics. For more details, please read our preprint here.
What's new?
- Nov. 24, 2020: Evan gave a great talk at MLCB 2020 about a zero-shot version of AMBER, tentatively under the name "AMBIENT". See his talk on YouTube here.
- Nov. 22, 2020: AMBER now has a logo. We also added a tutorial hosted by Google Colab.
AMBER is developed under Python 3.7 and Tensorflow 1.15.
Currently AMBER is designed to be run in Linux-like environment. As a prerequisite, please make sure you have Anaconda/miniconda installed, as we provide the detailed dependencies through a conda environment.
Please follow the steps below to install AMBER. There are two ways to install AMBER
: 1) cloning the latest development
from the GitHub repository and install with Anaconda
; and 2) using pypi
to install a versioned
release.
First, clone the Github Repository; if you have previous versions, make sure you pull the latest commits/changes:
git clone https://github.com/zj-zhang/AMBER.git
cd AMBER
git pull
If you see Already up to date
in your terminal, that means the code is at the latest change.
The easiest way to install AMBER is by Anaconda
. It is recommended to create a new conda
environment for AMBER:
conda create --file ./templates/conda_amber.linux_env.yml
python setup.py develop
As of version 0.1.0
, AMBER is on pypi. In the command-line terminal, type the following commands to get it installed:
pip install amber-automl
This will also install the required dependencies automatically. The pip install is still in its beta phase, so if you encouter any issues, please send me a bug report, and try installing with Anaconda as above.
You can test if AMBER can be imported to your new conda
environment like so:
conda activate amber
python -c "import amber"
If no errors pop up, that means you have successfully installed AMBER.
The typical installation process should take less than 10 minutes with regular network connection and hardware specifications.
The easist entry point to AMBER
is by following the tutorial
in Google colab, where you can run in a real-time, free GPU
environment.
- Tutorial is here: https://amber-automl.readthedocs.io/en/latest/resource/tutorials.html
- Open google colab notebook here.
In a nutshell, to run Amber
to build a Convolutional neural network, you will only need to provide the file
paths to compiled training and validation dataset, and specify the input and output shapes. The output of
AMBER is the optimal model architecture for this search run, and a full history of child models during architecture search.
Once you modified the string of file paths, The canonical way of triggering an AMBER run is simply:
from amber import Amber
# Here, define the types and specs using plain english in a dictionary
# You can refer to the examples under "template" folder
amb = Amber(types=type_dict, specs=specs)
amb.run()
Please refer to the template file for running transcriptional regulation prediction tasks using Convolutional Neural networks: here
Meanwhile, there are more if one would like to dig more. Going further, two levels of
settings are central to an Amber
instance: a types
dictionary and a specs
dictionary.
- The
types
dictionary will tellAmber
which types of components (such as controller and training environment) it should be using. - The
specs
will further detail every possible settings for thetypes
you specified. Only use this as an expert mode.
If you encounter any issues and/or would like feedbacks, please leave a GitHub issue. We will try to get back to you as soon as possible.