Defaults, eight months of daily data

match ← All linear growth benchmarks

256 days of history. Over 100 rows, so both libraries fit with L-BFGS instead of Newton.

Input

benchmark/inputs/v1/linear-medium.json: 256 history rows and 64 forecast rows.

Output

Largest difference: 0.0018 (0.0018% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	107.7958	107.7954	0.00045
2020-09-14	107.5631	107.5627	0.00045
2020-09-15	108.3425	108.3420	0.00047
2020-09-16	109.9333	109.9328	0.00052
2020-09-17	111.5460	111.5455	0.00054
2020-09-18	112.3359	112.3353	0.00055
2020-09-19	112.0778	112.0772	0.00057
2020-09-20	111.3662	111.3656	0.00060
2020-09-21	111.1336	111.1330	0.00059
2020-09-22	111.9129	111.9123	0.00062
2020-09-23	113.5037	113.5030	0.00067
2020-09-24	115.1164	115.1157	0.00069
2020-09-25	115.9063	115.9056	0.00070
2020-09-26	115.6482	115.6475	0.00072
2020-09-27	114.9366	114.9359	0.00075
2020-09-28	114.7040	114.7033	0.00074
2020-09-29	115.4833	115.4825	0.00077
2020-09-30	117.0741	117.0733	0.00082
2020-10-01	118.6868	118.6860	0.00083
2020-10-02	119.4767	119.4759	0.00085
2020-10-03	119.2186	119.2178	0.00087
2020-10-04	118.5071	118.5062	0.00090
2020-10-05	118.2744	118.2735	0.00089
2020-10-06	119.0537	119.0528	0.00092
2020-10-07	120.6445	120.6436	0.00096
2020-10-08	122.2573	122.2563	0.00098
2020-10-09	123.0471	123.0461	0.0010
2020-10-10	122.7890	122.7880	0.0010
2020-10-11	122.0775	122.0765	0.0010
2020-10-12	121.8448	121.8438	0.0010
2020-10-13	122.6242	122.6231	0.0011
2020-10-14	124.2150	124.2139	0.0011
2020-10-15	125.8277	125.8266	0.0011
2020-10-16	126.6176	126.6164	0.0011
2020-10-17	126.3595	126.3583	0.0012
2020-10-18	125.6479	125.6467	0.0012
2020-10-19	125.4153	125.4141	0.0012
2020-10-20	126.1946	126.1934	0.0012
2020-10-21	127.7854	127.7841	0.0013
2020-10-22	129.3981	129.3968	0.0013
2020-10-23	130.1880	130.1867	0.0013
2020-10-24	129.9299	129.9286	0.0013
2020-10-25	129.2183	129.2170	0.0013
2020-10-26	128.9857	128.9844	0.0013
2020-10-27	129.7650	129.7636	0.0014
2020-10-28	131.3558	131.3544	0.0014
2020-10-29	132.9685	132.9671	0.0014
2020-10-30	133.7584	133.7570	0.0014
2020-10-31	133.5003	133.4989	0.0015
2020-11-01	132.7888	132.7873	0.0015
2020-11-02	132.5561	132.5546	0.0015
2020-11-03	133.3354	133.3339	0.0015
2020-11-04	134.9262	134.9247	0.0016
2020-11-05	136.5390	136.5374	0.0016
2020-11-06	137.3288	137.3272	0.0016
2020-11-07	137.0707	137.0691	0.0016
2020-11-08	136.3592	136.3576	0.0016
2020-11-09	136.1265	136.1249	0.0016
2020-11-10	136.9058	136.9042	0.0017
2020-11-11	138.4967	138.4950	0.0017
2020-11-12	140.1094	140.1077	0.0017
2020-11-13	140.8993	140.8975	0.0017
2020-11-14	140.6412	140.6394	0.0018
2020-11-15	139.9296	139.9278	0.0018

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Convert input rows	40 µs	556 µs	14× faster
Fit	26.9 ms	87.4 ms	3.2× faster
Predict	316 µs	3.28 ms	10× faster
Fit and predict	26.2 ms	89.5 ms	3.4× faster
Convert, fit and predict	26.1 ms	91.3 ms	3.5× faster
First forecast in a new process	193 ms	1.51 s	7.8× faster
Save model to JSON	29 µs	583 µs	20× faster
Load model from JSON	157 µs	2.35 ms	15× faster
Load and predict in a new process	155 ms	1.44 s	9.3× faster

Settings

The case declaration both libraries ran.

{
  "kind": "map",
  "comparison": {
    "kind": "equivalent-objective",
    "evidenceId": "linear-output-first-v4"
  },
  "fitRequest": "growth-only",
  "configuration": {
    "growth": "linear",
    "scaling": "absmax",
    "seasonalityMode": "additive",
    "holidaysMode": "additive",
    "changepoints": {
      "mode": "auto",
      "count": 25,
      "range": 0.8
    },
    "changepointPriorScale": 0.05,
    "seasonalities": [
      {
        "name": "weekly",
        "periodDays": 7,
        "fourierOrder": 3,
        "priorScale": 10
      }
    ],
    "events": [],
    "regressors": []
  },
  "pythonOptimizer": {
    "algorithm": "Auto",
    "maxIterations": 10000,
    "newtonFallback": true
  }
}