create predictions from ARIMA and LSTM

README.md

@@ -18,7 +18,9 @@ Create model input data file from Yahoo Finance stock ticker API for particular
 
 Train multi variable time series model against input data and output prediction of values in the original problem space:
 
-`docker run --volume /local/path/to/api.json:/tmp/ml_timeseries/api.json --volume /local/path/to/prediction.json:/tmp/ml_timeseries/prediction.json docker.pkg.github.com/vraid-systems/ml_timeseries/main:2.0.0`
+`docker run --env MODEL=lstm --volume /local/path/to/api.json:/tmp/ml_timeseries/api.json --volume /local/path/to/prediction.json:/tmp/ml_timeseries/prediction.json docker.pkg.github.com/vraid-systems/ml_timeseries/main:2.0.0`
+
+Note that `MODEL` can be `arima` or `lstm`.
 
 
 Label the predicted features:

package.json

@@ -11,6 +11,7 @@
   },
   "dependencies": {
     "@tensorflow/tfjs-node": "^2.0.1",
+    "arima": "0.1.0",
     "array-smooth": "^1.0.0",
     "axios": "^0.19.2",
     "lodash": "^4.17.19",

src/MultiVariableArima.js

@@ -0,0 +1,76 @@
+const arima = require('arima')
+const lodash = require('lodash')
+
+class MultiVariableArima {
+    constructor(
+        ascendingHistoricalData,
+        barsToPredict = 4,
+    ) {
+        this.barsToPredict = barsToPredict
+        this.numberOfFeatureVariables = ascendingHistoricalData[0].length - 1
+        this.positionOfTimeFeature = 0
+
+        this.historicalData = ascendingHistoricalData.filter(
+            (currentElement) => (currentElement.length - 1) === this.numberOfFeatureVariables,
+        )
+        this.historicalFeatureValues = this.historicalData.map(
+            (currentFeatureTuple) => currentFeatureTuple.filter(
+                (
+                    currentElementInTuple, currentIndexInTuple,
+                ) => currentIndexInTuple !== this.positionOfTimeFeature,
+            ),
+        )
+    }
+
+    predictNextBars() {
+        const historicalTimeBars = this.historicalData.map(
+            (currentFeatureTuple) => currentFeatureTuple[this.positionOfTimeFeature],
+        )
+        const barSize = Math.abs(historicalTimeBars[1] - historicalTimeBars[0])
+        const mostRecentTimeBar = historicalTimeBars[historicalTimeBars.length - 1]
+
+        const getFeatureByPosition = (
+            historicalFeatureValues,
+            featurePosition,
+        ) => historicalFeatureValues.map(
+            (currentFeatureTuple) => currentFeatureTuple.filter(
+                (
+                    currentElementInTuple, currentIndexInTuple,
+                ) => currentIndexInTuple === featurePosition,
+            ),
+        )
+
+        const allFeaturePredictions = []
+        for (let index = 0; index < this.numberOfFeatureVariables; index += 1) {
+            const [singleFeaturePredictions] = arima(
+                getFeatureByPosition(this.historicalFeatureValues, index),
+                this.barsToPredict,
+                {
+                    auto: false,
+                    d: 0, // Number of times the series needs to be differenced
+                    method: 0, // ARIMA method (Default: 0)
+                    optimizer: 6, // Optimization method (Default: 6)
+                    p: 1, // Number of Autoregressive coefficients
+                    q: 1, // Number of Moving Average Coefficients
+                    transpose: false,
+                    verbose: true, // Output model analysis to console
+                },
+            )
+            allFeaturePredictions.push(singleFeaturePredictions)
+        }
+        const predictedFeatureTuples = lodash.zip(...allFeaturePredictions)
+
+        const predictedNextBars = []
+        for (let index = 0; index < this.barsToPredict; index += 1) {
+            const nextBarNumber = index + 1
+            predictedNextBars.push({
+                time: mostRecentTimeBar + (barSize * nextBarNumber),
+                features: predictedFeatureTuples[index],
+            })
+        }
+
+        return predictedNextBars
+    }
+}
+
+module.exports = MultiVariableArima