predictive-memory
v1.3.1
Published
Generates predictions and classifications using a hierarchical sequence/object cache
Downloads
2
Readme
predictive memory
predictive memory generates predictions and classifications using a hierarchical sequence/object cache.
Why
Prediction and classification can be implemented in low level neural networks (i.e. Numenta HTM). The goal of this module is to provide similar features, but do it faster while using less resources.
The long term goal is to expand and use this module in order to research cognition and perhaps consciousness.
How
This module provides a Predictor class that can process historical objects. It uses a Comparator to compare individual objects and sequences of those objects to other previously recorded sequences. The cache that is used to record sequences uses an internal hierarchy to quickly find similar key/value pairs in large amounts of historical data. In order to get balanced & weighted predictions, the second (or more) most similar cache result is used to weigh cache results whenever possible.
This module also provides support for text classification by providing a Classifier class. It uses the same type of hierarchical cache as the Predictor.
Installation
Via npm:
npm install predictive-memoryUsage
This module can be used both server and client side. Client side implementation can be done by loading predictive-memory.js as provided the root of the module. Please note that all class names in the client side implementation have been prefixed with 'Pm'.
Prediction
Server side example
See the example implementation in test/e2e/rec-center-hourly.js.
const { PredictorConfig, Predictor, PredictorAnalyzer } = require('predictive-memory');
const pc = new PredictorConfig();
const predictor = new Predictor(pc);
let prediction = null;
// Pass in any name and number of value properties
// Predictions will be generated when a minimal number of records have been processed
prediction = predictor.process({ dayOfWeek: 1, hourOfDay: 22, measuredValue: 5.5 });
prediction = predictor.process({ dayOfWeek: 1, hourOfDay: 23, measuredValue: 6.5 });
predictor.setPredict(true);
prediction = predictor.process({ dayOfWeek: 2, hourOfDay: 0, measuredValue: 1.5 });
// Optional methods to obtain predictor results (see PredictorConfig.maxPredictionHistorySize)
// const latestPrediction = predictor.predictions.get([0])[0];
// const results = predictor.getResults('measuredValue');
const analyzer = new PredictorAnalyzer();
const analysis = analyzer.analyze(predictor, ['measuredValue']);
console.log(analysis);
const predictionsForNext24Steps = predictor.generatePredictions(24);
console.log(predictionsForNext24Steps);Client side example
See the example implementation in test/e2e/test.html.
<html>
<head>
<script src="predictive-memory.js"></script>
<script type="text/javascript">
const onload = () => {
const pc = new PmPredictorConfig();
const predictor = new PmPredictor(pc);
for (let i = 0; i < 200; i++) {
predictor.process({k: i % 5, v: i % 10});
if (i > 10) {
predictor.setPredict(true);
}
}
const analyzer = new PmPredictorAnalyzer();
const analysis = analyzer.analyze(predictor, ['v']);
let str = JSON.stringify(analysis, null, 2);
str = str.replace(/ /g, ' ');
str = str.replace(/\n/g, '<br />');
document.getElementById('analysis').innerHTML = str;
};
</script>
</head>
<body onload="onload()">
<div id="analysis" />
</body>
</html>Classification
Server side example
See the example implementation in test/e2e/text-classification.js.
const { ClasifierConfig, Classifier, ClassifierAnalyzer } = require('predictive-memory');
const config = new ClassifierConfig();
const classifier = new Classifier(config);
// Record input for analyzer
classifier.setRecordInput(true);
classifier.put('String to associate1','Class to associate1');
classifier.put('String to associate2','Class to associate2');
const result classifier.classify('String to classify');
console.log(result.classifications);
const analyzer = new ClassifierAnalyzer();
const analysis = analyzer.analyze(classifier);
console.log(analysis);