predictive v0.0.3
predictive
Simple library for generating predictive, ebooks-esque text using word pairs. This began as a rewrite of nodeEbot, but dropped and added enough features to become its own thing.
Add to your project's package.json or install with npm install predictive
You could use it to make tweets:
var predictive = require('predictive');
var options = {
files: [
"path/to/text1.txt",
"path/to/text2.txt"
],
break_text_on: "\n",
lines_to_output: 1,
characters_per_line: 140,
select_next: "random"
}
predictive.generate(options,function(text){
console.log(text);
});Sample output, using Seinfeld and Game of Thrones scripts as text inputs (see examples folder):
JORAH: Ser Jorah Mormont of Bear Island. I served your father from his chess board.Or something bigger:
var predictive = require('predictive');
var options = {
files: [
"path/to/text1.txt",
"path/to/text2.txt"
],
break_text_on: "\n",
lines_to_output: 10,
sentences_per_line: 3,
select_next: "random"
}
predictive.generate(options,function(text){
console.log(text);
});Sample output, again using Seinfeld and Game of Thrones scripts as text inputs:
GEORGE: Why couldn't you have to help us here in New York Health Club]
RAMSAY: Why? Winterfell is yours.
SEPTA MORDANE: Enough!
EXT. SANSA’S ROOM - MORNING
ELAINE: No, you don't wear the ribbon. We are travelling together. In each other’s company.
SPARROW: A sinner comes before you with a squad of guardsmen to escort you to start a marriage. We are the lords of small matters here.
ELAINE: It isn't?
JERRY: But see look at them..
PHARMACIST: Did you make it up. That is why it is still laughing)
THEON: I’ve come to meet?Here are all the options
| Option | Type | Required | Default | Description |
|---|---|---|---|---|
files | Array | Yes | null | The relative path to the text to input. Supports txt (recommended), csv and json. I haven't done much testing with csv or json. Should work though! Let me know. |
break_text_on | String | For txt file(s) | "\n" | For an efficient and robust corpus, input text needs to be broken up into discrete lines. If you're inputting text with no line breaks, consider using punctuation here. |
csv_field | String | For csv file(s) | 0 | The field in your csv file from which text should be ingested. |
json_field | String | For json file(s) | 0 | The field in your json file from which text should be ingested. Your json input should probably be flat. |
lines_to_output | Integer | - | 10 | Number of lines of generated text to output. Line breaks will be appended to outputted text. |
characters_per_line | Integer | - | -1 (no limit) | Limit the number of characters for each line. Will often cutoff sentences at weird places. |
words_per_line | Integer | - | -1 (no limit) | Limit the number of words for each line. Will often cutoff sentences at weird places. |
sentences_per_line | Integer | - | -1 (no limit) | Number of sentences for each line. Sentences will be terminated by defined punctuation (see below). |
punctuation | Array | - | [".","!","?"] | Which characters should denote the end of a sentence. |
startwords | Array | - | - | An array of words to start each generated sentence with. Must be words that exist in corpus. If not defined, predictive will use the first word on each sentence in corpus. |
select_next | String | - | "random" | Method to use for selecting the next word in sentence generation. See options below. |
select_next options:
"random" (recommended) during line generation, selects the next word from an array of relevant word pairs at random.
"most_frequent" during line generation, selects the next word from an array of relevant word pairs by the word that occurs most frequently after current one.
"least_frequent" during line generation, selects the next word from an array of relevant word pairs by the word that occurs least frequently after current one.