A framework for training graph neural networks to untangle assembly graphs obtained from OLC-based de novo genome assemblers.
BSD 3-Clause "New" or "Revised" License
30
stars
3
forks
source link
Add decoding algorithms, command flags to supply strategy and parameters, hyperparameters to generalize the process and minor code quality improvements #5
The decoding (search) algorithms added are: depth d search, top k search, semi-random search, weighted random search (including random search to find a polynomial for it) and beam search, with parameters that can be specified using command flags.
The decoding process is also generalized, with customizable hyperparameters (applicable to all search algorithms) to change the heuristic values and the reduce function that combines individual heuristic values to give the heuristic value of a path.
Minor improvements are made to the code to reduce redundancy and improve readability (e.g. adding docstrings and abstracting backwards search with forwards search). With changes to the code, the naming of some variables has also been updated.
The decoding (search) algorithms added are: depth d search, top k search, semi-random search, weighted random search (including random search to find a polynomial for it) and beam search, with parameters that can be specified using command flags.
The decoding process is also generalized, with customizable hyperparameters (applicable to all search algorithms) to change the heuristic values and the reduce function that combines individual heuristic values to give the heuristic value of a path.
Minor improvements are made to the code to reduce redundancy and improve readability (e.g. adding docstrings and abstracting backwards search with forwards search). With changes to the code, the naming of some variables has also been updated.