DNADNA Documentation

Deep Neural Architecture for DNA.

The goal of this package is to provide utility functions to improve development of neural networks for population genetics.

dnadna should allow researchers to focus on their research project, be it the analysis of population genetic data or building new methods, without the need to focus on proper development methodology (unit test, continuous integration, documentation, etc.). Results will thus be more easily reproduced and shared. Having a common interface will also decrease the risk of bugs.

Full Documentation

Contents:

• Introduction
  • Installation
  • Dependencies
  • Quickstart Tutorial
  • Data format
  • Development
  • Detailed usage
• Usage Overview
  • Simulation
  • Initialization
  • Preprocessing
  • Training
  • Prediction
• Tutorials
  • Quickstart Tutorial 1
  • Quickstart Tutorial 2: Solve a classification task
  • Quickstart Tutorial 3: Train SPIDNA on an existing dataset
  • Quickstart Tutorial 4: Train AttentiveSPIDNA on an existing dataset
  • Predicting size fluctuations with a pre-trained network (Sanchez et al. 2020)
• Configuration Format
  • Configuration structure
  • Path resolution in config files
  • Configuration inheritance
  • Schemas
• Simulation Datasets
  • The DNADNA Dataset Format
  • Alternative dataset formats
• Dataset Preprocessing
  • What does preprocessing do?
  • How do you configure the preprocessing command?
  • What are the input and output files for the preprocessing step?
  • Command line
  • Next step
• Model Training
  • What does training do?
  • How do you configure the training command?
  • What are the output files for the training step?
  • Command line
  • Next step
• Prediction
  • What does prediction do?
  • How do you configure the predict command?
  • What are the output files for the predict step?
• Using and Implementing Simulators
  • Simulation config files
  • msprime Simulator Tutorial
• Configuration Schemas
  • Dataset config schema
  • Simulation config schema
  • Preprocessing config schema
  • Learned params schema
  • Training config schema
  • Summary statistics config schema
• Extending DNADNA with Plugins
  • Adding a network
• Development
  • Setup
  • Running the tests
  • Building and editing the docs
  • Releasing
• API Documentation
  • dnadna
  • dnadna.simulation
  • dnadna.data_preprocessing
  • dnadna.training
  • dnadna.nets
  • dnadna.transforms
  • dnadna.simulator
  • dnadna.snp_sample
  • dnadna.datasets
  • dnadna.params
  • dnadna.utils
  • dnadna.examples

Changelog

1.0.0rc2 (unreleased)

• Nothing changed yet

1.0.0rc1 (16-05-2022)

Enhancements

• Batch size can be customized when running dnadna predict (!140)

• SNP matrices are no longer forced to be unsigned 8-bit integers, but are allowed to be in principle any numeric data type that can converted to floats (!154)

• New syntax in config files for completely overriding inherited sections, rather than merging with them (!157)

Bug fixes

• When running dnadna predict on an existing file it is appended to without writing a new header (!134)

• Fixed bug in resolution of relative paths in config files when they are inherited from another config file (!135)

• Fixed bug where custom network parameters in the network configuration sometimes failed validation. Added better logging of inferred network params (!137)

• Fixed possible performance defect when running prediction due to acciental use of np.ext() on Tensor objects (!139)

• Fixed crash that could occur in dnadna preprocess when some replicates are missing from the dataset (!141)

• Fixed crash when running dnadna train on a purely classification task (!147)

• Fixed support for PyTorch 1.10+ (!153)

Misc

• The dnadna init command now requires a model name argument, and does not take a default name from the dataset name (!136)

• Added new dependency on jsonschema-pyref which allowed removing some of the more complicated and bug-prone code for validation of complex schemas (!144)

• Installation of tensorboard is now optional (!149)

• Various documentation improvements.

Indices and tables

• Index

• Module Index

• Search Page