ianbollinger/nomegen

a library and utility for randomly generating names

nomegen's Issues

Graphical user interface

It would be nice to have a GUI to go along with the CLI. The question is what toolkit to use. Currently, hsQML seems to be the most portable.

Release on Hackage

Prerequisites:

• finish documentation and README
• finish test suite
• finalize API

Binary serialization for Nomicons (and MarkovMaps)

Generating the MarkovMap is an expensive operation and is unnecessary if the underlying Nomicon has not changed. This is currently impossible to do, however, as the implementation of the data structure MarkovMap relies on is not exposed and provides no means of binary serialization.

Additionally, a binary format for Nomicon files would allow for faster deserialization, especially since individual names wouldn't need to be parsed into segments. For ease of use, nomegen would need to detect if the text format had been altered since it was last "compiled".

Write README

• Document the command line interface.
• Document the Nomicon file format.

Use Travis CI

Clean up lens-based interface

• Generate and export lenses for Nomicon and other data types.
• Don't export record fields or constructors.
• Rename and export isomorphisms.
• Simplify Cons/Snoc instances.

Add back configuration data type

We need a separate configuration type in addition to the Nomicon. It should store: the Nomicon, the size of the prediction context, and the generated MarkovMap.

Allow customizing the Markov chain context

Currently the value is hard-coded as 2 for "aesthetic" reasons.

Eliminate partial functions

Avoid calling partial functions like (!!) and incomplete pattern matches. Avoiding the latter will eliminate spurious warnings about (hypothetically) impossible matches. Additionally, don't call error when parsing fails; we need more context about the failure to provide a helpful error message anyway.

Refactor complex functions

• markovGenerate
• countSegments
• windows

YAML error messages

• Pretty-print YAML error messages.
• Attempt to provide a better error message than "mzero".
• Try to explain error messages, though we'd be better served just using a less complex format than YAML.
• Try not to leak the concrete details of YAML deserialization errors, though this has downsides.

The CLI should be able to read from stdin

But should this should be the default behavior?

Handle suffixes properly

Currently nomegen generates names that are as long as possible instead of potentially terminating on a suffix.

Unicode normalization

Currently we do no Unicode normalization on either names or segments, which means that seemingly identical characters could fail to parse unexpectedly.

Documentation

• Write top-level module documentation.
• Add description field to module header.
• Document complex algorithms: countSegments, windows, markovGenerate.
• Write description field in cabal file.

Replace YAML with custom format

Even YAML is too verbose a format and has weird quirks. Namely, having to quote the letters "n" and "y" for segments is confusing and forgetting to currently yields a meaningless error message. Eliminating yaml as a dependency may also eliminate a few transitive dependencies.

Prefix generation

Currently the generator merely selects a random segment to start with and then terminates after n segments are generated. The results are sloppy and we should instead always select the initial node in the Markov chain and terminate on the final node.

For instance, we could wrap each Segment in the MarkovMap in a data type like this:

data Component a
   = Initial
   | Medial !a
   | Final

Finish test suite

Provide example nomicon

The utility isn't too useful on its own without provided data to train it on.

Allow custom seeding for PRNG (consider modularizing PRNG)

There should at least be a backdoor to seed the PRNG. Additionally, it would be nice if the usage of the PRNG were modular and could be swapped out for another. Currently, mwc-random must be used for its generation of variates in a categorical distribution; however, we can't serialize the tables it generates. Thus it may have to be replaced anyway. Using something like Vose's Alias Method for generating variates seems optimal.

Migrate TODOs to issue tracker

The source code is currently littered with TODO comments that should be tracked here.