Two switchable weekly patterns

match flagged ← All linear growth benchmarks

256 daily rows. One weekly pattern applies on dense days, another on sparse days.

Input

benchmark/inputs/v1/map-conditional-seasonalities-medium.json: 256 history rows and 64 forecast rows.

Output

Largest difference: 0.0027 (0.0028% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	108.2021	108.2012	0.00090
2020-09-14	107.1673	107.1664	0.00091
2020-09-15	108.2711	108.2701	0.0010
2020-09-16	110.0093	110.0084	0.00094
2020-09-17	111.0944	111.0934	0.0010
2020-09-18	111.5652	111.5642	0.00094
2020-09-19	111.9707	111.9695	0.0011
2020-09-20	111.3706	111.3695	0.0011
2020-09-21	112.2118	112.2106	0.0011
2020-09-22	111.7796	111.7783	0.0012
2020-09-23	113.1091	113.1072	0.0019
2020-09-24	114.6028	114.6016	0.0012
2020-09-25	115.0736	115.0725	0.0011
2020-09-26	115.4791	115.4777	0.0013
2020-09-27	114.8790	114.8777	0.0013
2020-09-28	114.1842	114.1829	0.0013
2020-09-29	116.2214	116.2200	0.0014
2020-09-30	117.0262	117.0248	0.0013
2020-10-01	118.1113	118.1098	0.0014
2020-10-02	118.5820	118.5807	0.0013
2020-10-03	119.9595	119.9585	0.0010
2020-10-04	118.3874	118.3859	0.0015
2020-10-05	117.6926	117.6911	0.0015
2020-10-06	118.7964	118.7948	0.0016
2020-10-07	120.2310	120.2294	0.0016
2020-10-08	121.6197	121.6181	0.0016
2020-10-09	122.0905	122.0889	0.0015
2020-10-10	122.4959	122.4942	0.0017
2020-10-11	121.8959	121.8942	0.0017
2020-10-12	121.8381	121.8367	0.0014
2020-10-13	122.3049	122.3030	0.0018
2020-10-14	124.0430	124.0413	0.0017
2020-10-15	124.2407	124.2389	0.0018
2020-10-16	125.5989	125.5972	0.0017
2020-10-17	126.0044	126.0024	0.0019
2020-10-18	125.4043	125.4024	0.0019
2020-10-19	124.7095	124.7076	0.0019
2020-10-20	125.8133	125.8113	0.0020
2020-10-21	127.5515	127.5495	0.0019
2020-10-22	128.6366	128.6345	0.0020
2020-10-23	128.2503	128.2483	0.0020
2020-10-24	129.5128	129.5107	0.0021
2020-10-25	128.9127	128.9106	0.0021
2020-10-26	128.8549	128.8531	0.0018
2020-10-27	129.3217	129.3195	0.0022
2020-10-28	131.0599	131.0578	0.0021
2020-10-29	132.1450	132.1428	0.0022
2020-10-30	132.6158	132.6136	0.0021
2020-10-31	133.2319	133.2302	0.0017
2020-11-01	132.4212	132.4189	0.0023
2020-11-02	131.7263	131.7240	0.0023
2020-11-03	132.8301	132.8277	0.0024
2020-11-04	134.5683	134.5660	0.0023
2020-11-05	135.6534	135.6510	0.0024
2020-11-06	136.1242	136.1219	0.0023
2020-11-07	136.5297	136.5271	0.0025
2020-11-08	136.2696	136.2671	0.0025
2020-11-09	135.2347	135.2323	0.0025
2020-11-10	136.3386	136.3360	0.0026
2020-11-11	138.0768	138.0742	0.0025
2020-11-12	139.1619	139.1592	0.0026
2020-11-13	139.6326	139.6301	0.0025
2020-11-14	140.0381	140.0354	0.0027
2020-11-15	139.4380	139.4353	0.0027

Internal fit differences

The forecasts match, so this case passes. Both libraries search for the best-fitting parameters and stop when they are close enough. On this input they stopped at slightly different points, so these internal numbers differ by more than the threshold where we take a closer look.

Check	Effect Prophet	Python Prophet	Threshold
Stationarity residual (lower is closer to the best fit)	20.9828	6.1113	0.01

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Convert input rows	75 µs	617 µs	8.2× faster
Fit	3.69 ms	11.8 ms	3.2× faster
Predict	527 µs	3.90 ms	7.4× faster
Fit and predict	3.11 ms	15.7 ms	5.0× faster
Convert, fit and predict	2.60 ms	16.2 ms	6.2× faster
First forecast in a new process	174 ms	1.42 s	8.2× faster
Save model to JSON	20 µs	646 µs	32× faster
Load model from JSON	103 µs	2.70 ms	26× faster
Load and predict in a new process	158 ms	1.42 s	9.0× 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": [
        "2020-03-21T00:00:00.000Z",
        "2020-06-29T00:00:00.000Z"
      ]
    },
    "changepointPriorScale": 0.2,
    "seasonalities": [
      {
        "name": "weekly-dense",
        "periodDays": 7,
        "fourierOrder": 3,
        "priorScale": 10,
        "conditionName": "dense"
      },
      {
        "name": "weekly-sparse",
        "periodDays": 7,
        "fourierOrder": 3,
        "priorScale": 8,
        "conditionName": "sparse"
      }
    ],
    "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
  }
}