@radically-straightforward/examples v1.0.0

Radically Straightforward · Examples

🥸 Example data for testing

Installation

$ npm install @radically-straightforward/examples

Usage

import * as examples from "./index.mjs";

name()

export function name(): string;

Examples of names of people.

Sources

• First names: https://www.ssa.gov/oact/babynames/decades/century.html
• Last names: https://www.thoughtco.com/most-common-us-surnames-1422656

text()

export function text({
  model = textModel,
  length = 10,
}: {
  model?: {
    [predecessor: string]: {
      [successor: string]: {
        count: number;
        percentile: number;
      };
    };
  };
  length?: number;
} = {}): string;

Example text.

If the length is 0, then the text is short and may not contain punctuation, which is suitable, for example, for the title of a conversation.

The default model mostly talks about food. You may train your own model on other subjects by following these steps:

1. Create a file called urls.json with a list of Wikipedia articles on subjects that the model should talk about, for example:

   urls.json

   [
     "https://en.wikipedia.org/wiki/Maple_syrup",
     "https://en.wikipedia.org/wiki/Chocolate_chip"
   ]

2. Run the binary that comes with @radically-straightforward/examples to collect those Wikipedia articles:

   $ npx examples collect

3. A file called corpus.json is created with the collected Wikipedia articles, and the urls.json file is updated with more Wikipedia articles.

   Select the articles that you consider relevant in urls.json, return to step 1, and repeat until enough enough Wikipedia articles have been collected. A bigger corpus yields a richer model with more diverse example texts, but it also produces bigger files and risks going off-topic.

4. Train the model with the binary that comes with @radically-straightforward/examples:

   $ npx examples train

   This produces a file called model.json which includes the model, and its contents can be provided to text() as the model.

   At this point you may delete the files urls.json and corpus.json if you wish.

References

• https://healeycodes.com/generating-text-with-markov-chains
• https://www.youtube.com/watch?v=eGFJ8vugIWA

Avatars

The folder avatars/ contains 263 examples of faces.

These people don’t exist, the faces were collected from https://github.com/NVlabs/stylegan2 / https://drive.google.com/drive/folders/1mTeo3J3Jo6aYImBshLM6XRl_Ua8fqgVW and resized to 256×256px with avatars/index.mjs.

Related Work

Faker, casual, and so forth

These other libraries don’t include text that can be trained on arbitrary topics and avatars.