AtelierTomato.Markov.Model 1.0.0

AtelierTomato.Markov

This project offers Markov Chain logic, data structures, storage solutions, parsers, and renderers as libraries that can be used modularly.

Guide

This guide will offer a basic overview on how to set up an AtelierTomato.Markov implementation.

Parsers

The first step is to choose a parser or multiple parsers. The default SentenceParser, which can be found in AtelierTomato.Markov.Core can be used if you are handling a platform that has no markdown. It will separate punctuation and some contractions, focused specifically on the English language, this improves Markov chaining by allowing for more matches to occur where they otherwise would not. For instance, contractions like don't are separated into don 't when put in storage. This is especially useful for possessive possessive nouns like Kelsey's or Alice's, since they are separated out into Kelsey 's and Alice 's, the 's can be matched to any other instance of 's, improving sentence generation output. It is possible to skip using parsers, or to use your own, if you would prefer to though. Parsers for individual services can be found as packages named AtelierTomato.Markov.Service.X. All service-specific parsers extend off of the default SentenceParser. A current list of all parsers we have and what they offer is below:

• SentenceParser: Parses the English language in order to provide more efficient Markov chaining. Deletes links. Separates input into multiple sentences determined by punctuation marks or line breaks. Removes sentences that are smaller than a configurable MinimumInputLength.
• DiscordSentenceParser: Removes all Discord markdown. Removes or replaces certain bot prefixes (configurable). Removes sentences that begin with certain bot prefixes (configurable). Stores custom emojis in our own format (looks like this: e:EmojiName:). Completely ignores all text inside of code blocks. Removes discord server links, even those that do not begin with https://.

Sentence building

Before sentences can be stored in the database, they must be made into Sentences, as in, the type in AtelierTomato.Markov.Model. This type has 4 properties: IObjectOID OID, AuthorOID Author, DateTimeOffset Date, string Text. If you would like to write Sentences of type BookObjectOID to your storage solution, you must write the code to construct these Sentences yourself. However, we offer SentenceBuilders for the following, with explanations of what they need:

• DiscordSentenceBuilder with function Build(): Needs to be given an IGuild? and IChannel from Discord.Net.Core, ulong messageID, ulong userID, a DateTimeOffset, and an IEnumerable<string> featuring the text of the sentences parsed from the message, can also be given a different instance as a string over the default discord.com, for futureproofing if different instances of Discord ever exist.

Storage

To use AtelierTomato.Markov, you must use a storage implementation that implements our model for Sentence. We offer the interface ISentenceAccess that you may create implementations of if we do not have a storage solution that you prefer already offered. However, you must use something that extends from ISentenceAccess, and thus uses our model, as the Markov chain logic requires it. The base AtelierTomato.Markov.Storage package includes InMemorySentenceAccess as well as interfaces for Sentence and WordStatistic (more on this later) access. All other interface implementations are found in packages AtelierTomato.Markov.Storage.X. Current offered storage solutions:

• SqliteSentenceAccess
• InMemorySentenceAccess NOTE: Should only be used for testing, will likely be removed in the future. Stores Sentences in RAM, and they will be lost when the program is terminated. If you would like to use keyword generation (explained below), then the text that you store as a Sentence should also be stored using IWordStatisticAccess.WriteWordStatisticsFromString.

Keyword generation

This step can be skipped completely, however, keywords are powerful for improving sentence generation by allowing user input to nudge the chain to be "aware" of what the user is prompting the chain with. We offer the type WordStatistic, as well as the storage for it in IWordStatisticAccess and classes extending off of it. A WordStatistic is a simple measurement of how frequently a word appears. The current keyword generation logic, found in KeywordProvider, simply looks for the least common WordStatistic above a certain configurable MinimumAppearanceForKeyword in a given string. You must pass in a pre-parsed string if you are using the SentenceParser. There is also a configurable option to ignore certain words from being used as keywords. The function KeywordProvider.Find() will return a single word as a string that can then be used as the keyword in the generation step. Its functionality will be explained there.

Generation

If you do not want to use our provided keyword logic, which will be explained below, all you need to do is make an instance of MarkovChain, which must be provided options and an ISentenceAccess implementation, and run MarkovChain.Generate(). You must provide a SentenceFilter (found in AtelierTomato.Markov.Model), though this can be null. Sentence filtering, keywords, and first words can all be used in generation as parameters to Generate(). They are explained below:

• filter: Filters are of class SentenceFilter. This class includes the properties OID and Author, the same as that of the Sentence, these can be used to filter at any level, such as requiring a Sentence from the database to be from a certain Discord server or channel, or from a specific user.
• keyword: You can use this as a whitelist for Sentences when querying, the chain will first look for those that match the prevList that include that word, if it fails to find any, then it will fallback to searching everything else (restricted by filter still, for both).
• firstWord: This will force the chain to start with a certain word, even if the word does not exist in the database (this will cause it to just output that word).

Rendering

If you are using one of our SentenceParsers, you must use one of our SentenceRenderers, unless you are fine having output have random spaces in it. Similar to the parsers, the base SentenceRenderer is extended by all service-specific renderers, and can be used to render things to plain text, it is found in AtelierTomato.Markov.Core. Service-specific renderers are found in packages AtelierTomato.Markov.Service.X. All renderers are compatible with all parsers, if you want to parse input with one parser on one platform and render it with another on antoher platform, you can with no difficulties or oddities. A current list of all renderers we have and what they offer is below:

• SentenceRenderer: Renders separated punctuation and contractions to be connected again. Renders our stored custom emoji format into the name of the emoji.
• DiscordSentenceRenderer: Escapes all characters not found in 0-9a-zA-Z. Attempts to render our internal emoji format into a custom emoji that the implementation has access to that matches the name of the emoji, first checking the current Discord server and then all Discord servers it has access to.

You're done!

That is the full extent of what is necessary to generate sentences with this project.