Mixed effects, eight months, minmax scaling

match ← All flat growth benchmarks

The same features on 256 daily rows.

Input

benchmark/inputs/v1/flat-mixed-components-large.json: 256 history rows and 64 forecast rows.

Output

Largest difference: 4.0e-7 (< 0.0001% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	19.5305	19.5305	1.4e-8
2020-09-14	0.2957	0.2957	1.4e-7
2020-09-15	7.1696	7.1696	6.3e-8
2020-09-16	34.9807	34.9807	1.7e-7
2020-09-17	36.2008	36.2008	1.3e-7
2020-09-18	69.7415	69.7415	4.0e-7
2020-09-19	50.4458	50.4458	2.3e-7
2020-09-20	19.5305	19.5305	1.4e-8
2020-09-21	0.2957	0.2957	1.4e-7
2020-09-22	35.0006	35.0006	1.4e-7
2020-09-23	36.1809	36.1809	1.6e-7
2020-09-24	62.8406	62.8406	3.7e-7
2020-09-25	69.7415	69.7415	4.0e-7
2020-09-26	50.4458	50.4458	2.3e-7
2020-09-27	35.0006	35.0006	1.4e-7
2020-09-28	0.2957	0.2957	1.4e-7
2020-09-29	8.3698	8.3698	7.0e-8
2020-09-30	34.9807	34.9807	1.7e-7
2020-10-01	62.8406	62.8406	3.7e-7
2020-10-02	35.0006	35.0006	1.4e-7
2020-10-03	50.4458	50.4458	2.3e-7
2020-10-04	19.5305	19.5305	1.4e-8
2020-10-05	1.4958	1.4958	1.5e-7
2020-10-06	7.1696	7.1696	6.3e-8
2020-10-07	35.0006	35.0006	1.4e-7
2020-10-08	62.8406	62.8406	3.7e-7
2020-10-09	69.7415	69.7415	4.0e-7
2020-10-10	50.4458	50.4458	2.3e-7
2020-10-11	20.7307	20.7307	2.1e-8
2020-10-12	35.0006	35.0006	1.4e-7
2020-10-13	7.1696	7.1696	6.3e-8
2020-10-14	34.9807	34.9807	1.7e-7
2020-10-15	62.8406	62.8406	3.7e-7
2020-10-16	69.7415	69.7415	4.0e-7
2020-10-17	36.2008	36.2008	1.3e-7
2020-10-18	19.5305	19.5305	1.4e-8
2020-10-19	0.2957	0.2957	1.4e-7
2020-10-20	7.1696	7.1696	6.3e-8
2020-10-21	34.9807	34.9807	1.7e-7
2020-10-22	35.0006	35.0006	1.4e-7
2020-10-23	70.9417	70.9417	3.9e-7
2020-10-24	50.4458	50.4458	2.3e-7
2020-10-25	19.5305	19.5305	1.4e-8
2020-10-26	0.2957	0.2957	1.4e-7
2020-10-27	35.0006	35.0006	1.4e-7
2020-10-28	34.9807	34.9807	1.7e-7
2020-10-29	64.0408	64.0408	3.6e-7
2020-10-30	69.7415	69.7415	4.0e-7
2020-10-31	50.4458	50.4458	2.3e-7
2020-11-01	35.0006	35.0006	1.4e-7
2020-11-02	0.2957	0.2957	1.4e-7
2020-11-03	7.1696	7.1696	6.3e-8
2020-11-04	36.1809	36.1809	1.6e-7
2020-11-05	62.8406	62.8406	3.7e-7
2020-11-06	35.0006	35.0006	1.4e-7
2020-11-07	50.4458	50.4458	2.3e-7
2020-11-08	19.5305	19.5305	1.4e-8
2020-11-09	0.2957	0.2957	1.4e-7
2020-11-10	8.3698	8.3698	7.0e-8
2020-11-11	35.0006	35.0006	1.4e-7
2020-11-12	62.8406	62.8406	3.7e-7
2020-11-13	69.7415	69.7415	4.0e-7
2020-11-14	50.4458	50.4458	2.3e-7
2020-11-15	19.5305	19.5305	1.4e-8

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Convert input rows	67 µs	638 µs	9.5× faster
Fit	2.26 ms	9.13 ms	4.0× faster
Predict	515 µs	4.71 ms	9.2× faster
Fit and predict	2.23 ms	13.7 ms	6.2× faster
Convert, fit and predict	2.15 ms	14.5 ms	6.7× faster
First forecast in a new process	174 ms	1.44 s	8.2× faster
Save model to JSON	35 µs	819 µs	24× faster
Load model from JSON	194 µs	3.87 ms	20× faster
Load and predict in a new process	164 ms	1.44 s	8.8× faster

Settings

The case declaration both libraries ran.

{
  "kind": "map",
  "comparison": {
    "kind": "equivalent-objective",
    "evidenceId": "flat-map-mixed-v1"
  },
  "configuration": {
    "growth": "flat",
    "scaling": "minmax",
    "seasonalityMode": "additive",
    "holidaysMode": "additive",
    "changepoints": {
      "mode": "auto",
      "count": 25,
      "range": 0.8
    },
    "changepointPriorScale": 0.05,
    "seasonalities": [
      {
        "name": "weekly-custom",
        "periodDays": 7,
        "fourierOrder": 2,
        "priorScale": 10,
        "conditionName": "active",
        "mode": "multiplicative"
      }
    ],
    "events": [
      {
        "name": "campaign",
        "date": "2020-01-22",
        "lowerWindowDays": 0,
        "upperWindowDays": 0,
        "priorScale": 10
      }
    ],
    "regressors": [
      {
        "name": "promotion",
        "priorScale": 10,
        "standardization": "never"
      }
    ]
  },
  "effectOptimizer": {
    "algorithm": "lbfgs",
    "maxIterations": 10000,
    "fallback": "none",
    "lbfgs": {
      "historySize": 5,
      "initAlpha": 0.001,
      "tolObj": 1e-12,
      "tolRelObj": 10000,
      "tolGrad": 1e-8,
      "tolRelGrad": 10000000,
      "tolParam": 1e-8
    }
  },
  "pythonOptimizer": {
    "algorithm": "LBFGS",
    "maxIterations": 10000,
    "newtonFallback": false
  }
}