ts-arima-forecast

import { Stepwise, RollingWindow, Adaptive } from '../../utils/strategies'; import { ARIMA } from '../../models/arima'; describe('Forecasting Strategies', () => { let baseModel: ARIMA; let trainData: number[]; let testData: number[]; const btcData = [ 67766.85, 67765.63, 68809.9, 70537.84, 71108, 70799.06, 69355.6, 69310.46, 69648.14, 69540, 67314.24, 68263.99, 66773.01, 66043.99, 66228.25, 66676.87, 66504.33, 65175.32, 64974.37, 64869.99, 64143.56, 64262.01, 63210.01, 60293.3, 61806.01, 60864.99, 61706.47, 60427.84, 60986.68, 62772.01, 62899.99, 62135.47, 60208.58, 57050.01, 56628.79, 58230.13, 55857.81, 56714.62, 58050, 57725.85, 57339.89, 57889.1, 59204.02, 60797.91, 64724.14, 65043.99, 64087.99, 63987.92, 66660, 67139.96, 68165.34, 67532.01, 65936.01, 65376, 65799.95, 67907.99, 67896.5, 68249.88, 66784.69, 66188, 64628, 65354.02, 61498.33, 60697.99, 58161, 54018.81, 56022.01, 55134.16, 61685.99, 60837.99, 60923.51, 58712.59, 59346.64, 60587.15, 58683.39, 57541.06, 58874.6, 59491.99, 58427.35, 59438.5, 59013.8, 61156.03, 60375.84, 64037.24, 64157.01, 64220, 62834, 59415, 59034.9, 59359.01, 59123.99, 58973.99, 57301.86, 59132.13, 57487.73, 57970.9, 56180, 53962.97, 54160.86, 54869.95, 57042, 57635.99, 57338, 58132.32, 60498, 59993.03, 59132, 58213.99, 60313.99, 61759.99, 62947.99, 63201.05, 63348.96, 63578.76, 63339.99, 64262.7, 63152.01, 65173.99, 65769.95, 65858, 65602.01, 63327.59, 60805.78, 60649.28, 60752.71, 62086, 62058, 62819.91, 62224, 62160.49, 60636.02, 60326.39, 62540, 63206.22, 62870.02, 66083.99, 67074.14, 67620.01, 67421.78, 68428, 68378, 69031.99, 67377.5, 67426, 66668.65, 68198.28, 66698.33, 67092.76, 68021.7, 69962.21, 72736.42, 72344.74, 70292.01, 69496.01, 69374.74, 68775.99, 67850.01, 69372.01, 75571.99, 75857.89, 76509.78, 76677.46, 80370.01, 88647.99, 87952.01, 90375.2, 87325.59, 91032.07, 90586.92, 89855.99, 90464.08, 92310.79, 94286.56, 98317.12, 98892, 97672.4, 97900.04, 93010.01, 91965.16, 95863.11, 95643.98, 97460, 96407.99, 97185.18, 95840.62, 95849.69, 98587.32, 96945.63, 99740.84, 99831.99, 101109.59, 97276.47, 96593, 101125, 100004.29, 101424.25, 101420, 104463.99, 106058.66, 106133.74, 100204.01, 97461.86, 97805.44, 97291.99, 95186.27, 94881.47, 98663.58, 99429.6, 95791.6, 94299.03, 95300, 93738.2, 92792.05, 93576, 94591.79, 96984.79, 98174.18, 98220.5, 98363.61, 102235.6, 96954.61, 95060.61, 92552.49, 94726.11, 94599.99, 94545.06, 94536.1, 96560.86, 100497.35, 99987.3, 104077.48, 104556.23, 101331.57, 102260.01, 106143.82, 103706.66, 103910.34, 104870.5, 104746.85, 102620, 102082.83, 101335.52, 103733.24, 104722.94, 102429.56, 100635.65, 97700.59, 101328.52, 97763.13, 96612.43, 96554.35, 96506.8, 96444.74, 96462.75, 97430.82, 95778.2, 97869.99, 96608.14, 97500.48, 97569.66, 96118.12, 95780, 95671.74, 96644.37, 98305, 96181.98, 96551.01, 96258, 91552.88, 88680.4, 84250.09, 84708.58, 84349.94, 86064.53, 94270, 86220.61, 87281.98, 90606.01, 89931.89, 86801.75, 86222.45, 80734.37, 78595.86, 82932.99, 83680.12, 81115.78, 83983.2, 84338.44, 82574.53, 84010.03, 82715.03, 86845.94, 84223.39, 84088.79, 83840.59, 86082.5, 87498.16, 87392.87, 86909.17, 87232.01, 84424.38, 82648.54, 82389.99, 82550.01, 85158.34, 82516.29, 83213.09, 83889.87, 83537.99, 78430, 79163.24, 76322.42, 82615.22, 79607.3, 83423.84, 85276.9, 83760, 84591.58, 83643.99, 84030.38, 84947.91, 84474.69, 85077.01, 85179.24, 87516.23, 93442.99, 93691.08, 93980.47, 94638.68, 94628, 93749.3, 95011.18, 94256.82, 94172, 96489.91, 96887.14, 95856.42, 94277.62, 94733.68, 96834.02, 97030.5, 103261.6, 102971.99, 104809.53, 104118, 102791.32, 104103.72, 103507.82, 103763.71, 103463.9, 103126.65, 106454.26, 105573.74, 106849.99, 109643.99, 111696.21, 107318.3, 107761.91, 109004.19, 109434.79, 108938.17, 107781.78, 105589.75, 103985.48, 104644 ] beforeEach(() => { // Generate consistent test data trainData = btcData.slice(btcData.length * 0.8); testData = btcData.slice(0, btcData.length * 0.2); // Create and fit base model baseModel = new ARIMA({ p: 1, d: 1, q: 1 }); baseModel.fit(trainData); }); describe('Stepwise Strategy', () => { let stepwise: Stepwise; beforeEach(() => { stepwise = new Stepwise(baseModel, trainData, { refitModel: true, verbose: false }); }); test('should initialize correctly', () => { expect(stepwise).toBeInstanceOf(Stepwise); expect(stepwise.getCurrentData()).toEqual(trainData); expect(stepwise.getCurrentModel()).toBeInstanceOf(ARIMA); }); test('should forecast multiple steps ahead', () => { const result = stepwise.forecast(5); expect(result.forecasts).toHaveLength(5); expect(result.confidenceIntervals.lower).toHaveLength(5); expect(result.confidenceIntervals.upper).toHaveLength(5); // Check that all values are finite numbers result.forecasts.forEach(forecast => { expect(Number.isFinite(forecast)).toBe(true); }); result.confidenceIntervals.lower.forEach(lower => { expect(Number.isFinite(lower)).toBe(true); }); result.confidenceIntervals.upper.forEach(upper => { expect(Number.isFinite(upper)).toBe(true); }); }); test('should process real-time data batch', () => { const result = stepwise.forecastWithRealTimeData(testData); expect(result.forecasts).toHaveLength(testData.length); expect(result.errors).toHaveLength(testData.length); expect(result.actualValues).toEqual(testData); expect(result.updatedModels).toHaveLength(testData.length); // Check that all errors are non-negative result.errors!.forEach(error => { expect(error).toBeGreaterThanOrEqual(0); }); }); test('should add single observation and forecast', () => { const newObservation = 100.5; const result = stepwise.addObservationAndForecast(newObservation); expect(result.forecast).toBeDefined(); expect(Number.isFinite(result.forecast)).toBe(true); expect(Number.isFinite(result.lowerBound)).toBe(true); expect(Number.isFinite(result.upperBound)).toBe(true); expect(result.error).toBeUndefined(); // First call should have no error // Second call should have error const result2 = stepwise.addObservationAndForecast(101.2); expect(result2.error).toBeDefined(); expect(result2.error).toBeGreaterThanOrEqual(0); }); test('should update internal state correctly', () => { const initialLength = stepwise.getCurrentData().length; const newObservation = 100.5; stepwise.addObservationAndForecast(newObservation); expect(stepwise.getCurrentData()).toHaveLength(initialLength + 1); expect(stepwise.getCurrentData()[initialLength]).toBe(newObservation); }); test('should reset correctly', () => { const newData = [1, 2, 3, 4, 5]; stepwise.reset(newData); expect(stepwise.getCurrentData()).toEqual(newData); }); test('should work without model refitting', () => { const stepwiseNoRefit = new Stepwise(baseModel, trainData, { refitModel: false, verbose: false }); const result = stepwiseNoRefit.forecastWithRealTimeData(testData.slice(0, 3)); expect(result.updatedModels).toHaveLength(0); // No models updated when refitting is disabled }); }); describe('RollingWindow Strategy', () => { let rollingWindow: RollingWindow; const windowSize = 20; beforeEach(() => { rollingWindow = new RollingWindow(baseModel, trainData, { windowSize, verbose: false }); }); test('should initialize correctly', () => { expect(rollingWindow).toBeInstanceOf(RollingWindow); expect(rollingWindow.getWindowSize()).toBe(windowSize); expect(rollingWindow.getCurrentData()).toEqual(trainData); }); test('should forecast multiple steps ahead', () => { const result = rollingWindow.forecast(5); expect(result.forecasts).toHaveLength(5); expect(result.confidenceIntervals.lower).toHaveLength(5); expect(result.confidenceIntervals.upper).toHaveLength(5); }); test('should process real-time data batch', () => { const result = rollingWindow.forecastWithRealTimeData(testData); console.log(result); expect(result.forecasts).toHaveLength(testData.length); expect(result.errors).toHaveLength(testData.length); expect(result.actualValues).toEqual(testData); // Check that all errors are non-negative result.errors!.forEach(error => { expect(error).toBeGreaterThanOrEqual(0); }); }); test('should maintain rolling window size', () => { // Add more observations than window size const manyObservations = Array.from({ length: windowSize + 10 }, (_, i) => 100 + i); for (const obs of manyObservations) { rollingWindow.addObservationAndForecast(obs); } // Data should not exceed initial length + window size expect(rollingWindow.getCurrentData().length).toBeLessThanOrEqual(trainData.length + windowSize + 10); }); test('should add single observation and forecast', () => { const newObservation = 100.5; const result = rollingWindow.addObservationAndForecast(newObservation); expect(result.forecast).toBeDefined(); expect(Number.isFinite(result.forecast)).toBe(true); expect(Number.isFinite(result.lowerBound)).toBe(true); expect(Number.isFinite(result.upperBound)).toBe(true); expect(result.error).toBeUndefined(); // First call should have no error // Second call should have error const result2 = rollingWindow.addObservationAndForecast(101.2); expect(result2.error).toBeDefined(); expect(result2.error).toBeGreaterThanOrEqual(0); }); test('should allow window size adjustment', () => { const newWindowSize = 15; rollingWindow.setWindowSize(newWindowSize); expect(rollingWindow.getWindowSize()).toBe(newWindowSize); }); test('should reset correctly', () => { const newData = [1, 2, 3, 4, 5]; rollingWindow.reset(newData); expect(rollingWindow.getCurrentData()).toEqual(newData); }); test('should handle small window sizes', () => { const smallWindow = new RollingWindow(baseModel, trainData, { windowSize: 5, verbose: false }); const result = smallWindow.forecast(3); expect(result.forecasts).toHaveLength(3); }); }); describe('Adaptive Strategy', () => { let adaptive: Adaptive; beforeEach(() => { adaptive = new Adaptive(baseModel, trainData, { windowSize: 20, adaptationThreshold: 2.0, maxErrorWindowSize: 5, verbose: false }); }); test('should initialize correctly', () => { expect(adaptive).toBeInstanceOf(Adaptive); expect(adaptive.getCurrentStrategy()).toBe('stepwise'); // Should start with stepwise }); test('should forecast multiple steps ahead', () => { const result = adaptive.forecast(5); expect(result.forecasts).toHaveLength(5); expect(result.confidenceIntervals.lower).toHaveLength(5); expect(result.confidenceIntervals.upper).toHaveLength(5); }); test('should process real-time data batch', () => { const result = adaptive.forecastWithRealTimeData(testData); expect(result.forecasts).toHaveLength(testData.length); expect(result.errors).toHaveLength(testData.length); expect(result.actualValues).toEqual(testData); }); test('should add single observation and forecast', () => { const newObservation = 100.5; const result = adaptive.addObservationAndForecast(newObservation); expect(result.forecast).toBeDefined(); expect(Number.isFinite(result.forecast)).toBe(true); expect(Number.isFinite(result.lowerBound)).toBe(true); expect(Number.isFinite(result.upperBound)).toBe(true); expect(result.error).toBeUndefined(); // First call should have no error }); test('should switch strategies based on performance', () => { // Create data that should trigger strategy switching const highErrorData = Array.from({ length: 10 }, () => Math.random() * 1000); const initialStrategy = adaptive.getCurrentStrategy(); adaptive.forecastWithRealTimeData(highErrorData); // Strategy might have switched due to high errors const finalStrategy = adaptive.getCurrentStrategy(); expect(['stepwise', 'rolling']).toContain(finalStrategy); }); test('should reset correctly', () => { const newData = [1, 2, 3, 4, 5]; adaptive.reset(newData); expect(adaptive.getCurrentStrategy()).toBe('stepwise'); // Should reset to stepwise }); test('should handle strategy switching with verbose logging', () => { const verboseAdaptive = new Adaptive(baseModel, trainData, { windowSize: 20, adaptationThreshold: 1.0, // Low threshold to trigger switching maxErrorWindowSize: 3, verbose: true }); // Add observations that might trigger switching for (let i = 0; i < 5; i++) { verboseAdaptive.addObservationAndForecast(Math.random() * 100); } expect(['stepwise', 'rolling']).toContain(verboseAdaptive.getCurrentStrategy()); }); }); describe('Strategy Comparison', () => { test('should compare all strategies on same data', () => { const strategies = { stepwise: new Stepwise(baseModel, trainData, { refitModel: true }), rollingWindow: new RollingWindow(baseModel, trainData, { windowSize: 20 }), adaptive: new Adaptive(baseModel, trainData, { windowSize: 20 }) }; const results: { [key: string]: number } = {}; for (const [name, strategy] of Object.entries(strategies)) { const result = strategy.forecastWithRealTimeData(testData); const avgError = result.errors!.reduce((a, b) => a + b, 0) / result.errors!.length; results[name] = avgError; expect(avgError).toBeGreaterThanOrEqual(0); expect(result.forecasts).toHaveLength(testData.length); } // All strategies should produce results expect(Object.keys(results)).toHaveLength(3); expect(results.stepwise).toBeDefined(); expect(results.rollingWindow).toBeDefined(); expect(results.adaptive).toBeDefined(); }); }); describe('Edge Cases', () => { test('should handle empty test data', () => { const stepwise = new Stepwise(baseModel, trainData); const result = stepwise.forecastWithRealTimeData([]); expect(result.forecasts).toHaveLength(0); expect(result.errors).toHaveLength(0); expect(result.actualValues).toHaveLength(0); }); test('should handle single observation', () => { const rollingWindow = new RollingWindow(baseModel, trainData); const result = rollingWindow.forecastWithRealTimeData([100.5]); expect(result.forecasts).toHaveLength(1); expect(result.errors).toHaveLength(1); }); test('should handle very small training data', () => { const smallData = [1, 2, 3]; const smallModel = new ARIMA({ p: 1, d: 0, q: 1 }); smallModel.fit(smallData); const stepwise = new Stepwise(smallModel, smallData); const result = stepwise.forecast(2); expect(result.forecasts).toHaveLength(2); }); test('should handle large forecast horizons', () => { const stepwise = new Stepwise(baseModel, trainData); const result = stepwise.forecast(50); expect(result.forecasts).toHaveLength(50); expect(result.confidenceIntervals.lower).toHaveLength(50); expect(result.confidenceIntervals.upper).toHaveLength(50); }); }); });