Weekly pattern on irregular dates

match ← All linear growth benchmarks

192 unevenly spaced rows with a custom weekly seasonality and two changepoints.

Input

benchmark/inputs/v1/map-irregular-medium.json: 192 history rows and 48 forecast rows.

Output

Largest difference: 6.9e-5 (< 0.0001% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-08-17	84.4721	84.4720	3.6e-5
2020-08-18	84.7457	84.7457	1.9e-5
2020-08-19	85.3132	85.3132	3.1e-5
2020-08-20	85.8875	85.8875	4.4e-5
2020-08-22	86.6631	86.6631	4.8e-5
2020-08-23	87.0718	87.0718	3.5e-5
2020-08-24	87.4771	87.4770	4.0e-5
2020-08-25	87.7507	87.7507	2.2e-5
2020-08-26	88.3182	88.3182	3.5e-5
2020-08-27	88.8925	88.8925	4.8e-5
2020-08-28	89.1471	89.1470	4.6e-5
2020-08-30	90.0768	90.0768	3.9e-5
2020-08-31	90.4820	90.4820	4.3e-5
2020-09-01	90.7557	90.7556	2.6e-5
2020-09-02	91.3232	91.3231	3.8e-5
2020-09-03	91.8975	91.8974	5.1e-5
2020-09-04	92.1521	92.1520	4.9e-5
2020-09-06	93.0818	93.0818	4.2e-5
2020-09-08	93.7606	93.7606	2.9e-5
2020-09-09	94.3282	94.3281	4.2e-5
2020-09-10	94.9025	94.9024	5.5e-5
2020-09-11	95.1570	95.1570	5.3e-5
2020-09-12	95.6780	95.6780	5.8e-5
2020-09-13	96.0868	96.0867	4.6e-5
2020-09-14	96.4920	96.4919	5.0e-5
2020-09-16	97.3331	97.3331	4.5e-5
2020-09-17	97.9074	97.9074	5.8e-5
2020-09-18	98.1620	98.1620	5.6e-5
2020-09-19	98.6830	98.6829	6.2e-5
2020-09-20	99.0917	99.0917	4.9e-5
2020-09-21	99.4970	99.4969	5.4e-5
2020-09-22	99.7706	99.7706	3.6e-5
2020-09-24	100.9124	100.9124	6.2e-5
2020-09-25	101.1670	101.1669	6.0e-5
2020-09-26	101.6880	101.6879	6.5e-5
2020-09-27	102.0967	102.0967	5.3e-5
2020-09-29	102.7756	102.7755	4.0e-5
2020-09-30	103.3431	103.3430	5.2e-5
2020-10-01	103.9174	103.9173	6.5e-5
2020-10-03	104.6930	104.6929	6.9e-5
2020-10-04	105.1017	105.1016	5.6e-5
2020-10-05	105.5069	105.5069	6.1e-5
2020-10-06	105.7806	105.7805	4.3e-5
2020-10-07	106.3481	106.3480	5.6e-5
2020-10-08	106.9224	106.9223	6.9e-5
2020-10-09	107.1769	107.1769	6.7e-5
2020-10-11	108.1067	108.1066	6.0e-5
2020-10-12	108.5119	108.5118	6.4e-5

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Convert input rows	30 µs	533 µs	17× faster
Fit	2.00 ms	8.05 ms	4.0× faster
Predict	399 µs	3.17 ms	7.9× faster
Fit and predict	1.49 ms	11.3 ms	7.6× faster
Convert, fit and predict	1.57 ms	11.6 ms	7.4× faster
Save model to JSON	17 µs	515 µs	30× faster
Load model from JSON	95 µs	2.35 ms	25× faster
Load and predict in a new process	165 ms	1.42 s	8.6× faster

Settings

The case declaration both libraries ran.

{
  "kind": "linear-map",
  "comparison": {
    "kind": "equivalent-objective",
    "evidenceId": "linear-output-first-v4"
  },
  "configuration": {
    "effectMap": "explicit",
    "changepoints": {
      "mode": "explicit",
      "timestamps": [
        "2020-03-26T00:00:00.000Z",
        "2020-06-09T00:00:00.000Z"
      ]
    },
    "changepointPriorScale": 0.2,
    "seasonalities": [
      {
        "name": "weekly-custom",
        "periodDays": 7,
        "fourierOrder": 3,
        "priorScale": 10
      }
    ],
    "events": [],
    "regressors": []
  },
  "effectOptimizer": {
    "algorithm": "auto",
    "maxIterations": 10000,
    "fallback": "newton",
    "lbfgs": {
      "historySize": 5,
      "initAlpha": 0.001,
      "tolObj": 1e-12,
      "tolRelObj": 10000,
      "tolGrad": 1e-8,
      "tolRelGrad": 10000000,
      "tolParam": 1e-8
    }
  },
  "pythonOptimizer": {
    "algorithm": "Auto",
    "maxIterations": 10000,
    "newtonFallback": true,
    "sigFigs": 12
  }
}