Refit after a week of new data

match ← All flat growth benchmarks

fit(history, { growth: "flat", map: { warmStart } }) from warmStartParameters of a fit of the first 89 days, against fit(df, init=warm_start_params(m)). 96 days around one level with a weekly rhythm, fitted with Newton. Timings compare the refit with a fresh fit.

Input

benchmark/inputs/v1/flat-seasonal.json: 96 history rows and 24 forecast rows.

Output

Largest difference: 5.6e-6 (0.00019% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-04-06	33.6334	33.6334	1.4e-6
2020-04-07	33.9115	33.9115	3.2e-6
2020-04-08	35.0025	35.0025	4.0e-6
2020-04-09	36.0968	36.0968	5.6e-6
2020-04-10	36.3663	36.3663	2.5e-6
2020-04-11	35.6078	35.6078	2.8e-6
2020-04-12	34.3921	34.3921	3.6e-6
2020-04-13	33.6334	33.6334	1.4e-6
2020-04-14	33.9115	33.9115	3.2e-6
2020-04-15	35.0025	35.0025	4.0e-6
2020-04-16	36.0968	36.0968	5.6e-6
2020-04-17	36.3663	36.3663	2.5e-6
2020-04-18	35.6078	35.6078	2.8e-6
2020-04-19	34.3921	34.3921	3.6e-6
2020-04-20	33.6334	33.6334	1.4e-6
2020-04-21	33.9115	33.9115	3.2e-6
2020-04-22	35.0025	35.0025	4.0e-6
2020-04-23	36.0968	36.0968	5.6e-6
2020-04-24	36.3663	36.3663	2.5e-6
2020-04-25	35.6078	35.6078	2.8e-6
2020-04-26	34.3921	34.3921	3.6e-6
2020-04-27	33.6334	33.6334	1.4e-6
2020-04-28	33.9115	33.9115	3.2e-6
2020-04-29	35.0025	35.0025	4.0e-6

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Fit	110 ms	389 ms	3.5× faster
Refit from an earlier fit	6.38 ms	8.52 ms	1.3× faster

Settings

The case declaration both libraries ran.

{
  "kind": "map",
  "comparison": {
    "kind": "equivalent-objective",
    "evidenceId": "flat-growth-reduced-v1"
  },
  "fitRequest": "growth-only",
  "configuration": {
    "growth": "flat",
    "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
  },
  "warmStart": {
    "baseRows": 89,
    "investigateParameterAbsolute": 0.01
  }
}