avsecz / concise

CONCISE (COnvolutional neural Networks for CIS-regulatory Elements)

Concise: Keras extension for regulatory genomics

Concise (originally CONvolutional neural networks for CIS-regulatory Elements) allows you to:

1. Pre-process sequence-related data (concise.preprocessing)
   • convert a list of sequences into one-hot-encoded numpy array or tokens.
2. Specify a Keras model with additional modules
   • Concise provides custom layers, initializers and regularizers.
3. Tune the hyper-parameters (concise.hyopt)
   • Concise provides convenience functions for working with the hyperopt package.
4. Interpret the model
   • most of Concise layers contain plotting methods
5. Share and re-use models
   • every component (layer, initializer, regularizer, loss) is fully compatible with Keras. Model saving and loading works out-of-the-box.

Installation

Concise is available for Python versions greater than 3.4 and can be installed from PyPI using pip:

pip install concise

To successfully use concise plotting functionality, please also install the libgeos library required by the shapely package:

• Ubuntu: sudo apt-get install -y libgeos-dev
• Red-hat/CentOS: sudo yum install geos-devel

Documentation

• https://i12g-gagneurweb.in.tum.de/public/docs/concise/

concise's Issues

feature: multi-task learning

• w_motif -> matrix
• b -> vector
• optional w_feat -> matrix
• y -> matrix

add method to ConciseCV - get_performance(type = "mse") or similar

type = "mse", "r-squared", "variance explained"

Update sklearn/cross_validation.py to model_selection

sklearn/cross_validation.py:44: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.
"This module will be removed in 0.20.", DeprecationWarning)

feature: remove regress_out_feat feature

fix: readthedocs build

Ranking Metrics

I think it's good to have these in concise: https://gist.github.com/nkhuyu/ab5b0b29d55bcd6b80b8f11c2933636c

update tensorflow compatibility

WARNING:tensorflow:tf.op_scope(values, name, default_name) is deprecated, use tf.name_scope(name, default_name, values) _build_graph.: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02. Usetf.global_variables_initializerinstead.

feature: conciseCV global model - train for optimal number of epochs

At each fold, store the number of optimal epochs and train the global model for so many

fix: don't install tensorflow if tensorflow-gpu is installed

Asked in this thread:
http://stackoverflow.com/questions/41956902/how-to-declare-or-dependence-in-setup-py