Mixed effects, eight months, absmax scaling

match ← All flat growth benchmarks

The same features on 256 daily rows with absmax scaling.

Input

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

Output

Largest difference: 1.9e-7 (< 0.0001% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	28.9196	28.9196	1.6e-7
2020-09-14	21.3587	21.3587	1.9e-7
2020-09-15	24.0607	24.0607	1.5e-7
2020-09-16	34.9928	34.9928	5.1e-8
2020-09-17	36.2008	36.2008	7.4e-8
2020-09-18	48.6567	48.6567	1.5e-8
2020-09-19	41.0719	41.0719	6.4e-8
2020-09-20	28.9196	28.9196	1.6e-7
2020-09-21	21.3587	21.3587	1.9e-7
2020-09-22	35.0006	35.0006	8.2e-8
2020-09-23	36.1930	36.1930	4.3e-8
2020-09-24	45.9441	45.9441	2.1e-8
2020-09-25	48.6567	48.6567	1.5e-8
2020-09-26	41.0719	41.0719	6.4e-8
2020-09-27	35.0006	35.0006	8.2e-8
2020-09-28	21.3587	21.3587	1.9e-7
2020-09-29	25.2609	25.2609	1.4e-7
2020-09-30	34.9928	34.9928	5.1e-8
2020-10-01	45.9441	45.9441	2.1e-8
2020-10-02	35.0006	35.0006	8.2e-8
2020-10-03	41.0719	41.0719	6.4e-8
2020-10-04	28.9196	28.9196	1.6e-7
2020-10-05	22.5589	22.5589	1.9e-7
2020-10-06	24.0607	24.0607	1.5e-7
2020-10-07	35.0006	35.0006	8.2e-8
2020-10-08	45.9441	45.9441	2.1e-8
2020-10-09	48.6567	48.6567	1.5e-8
2020-10-10	41.0719	41.0719	6.4e-8
2020-10-11	30.1197	30.1197	1.5e-7
2020-10-12	35.0006	35.0006	8.2e-8
2020-10-13	24.0607	24.0607	1.5e-7
2020-10-14	34.9928	34.9928	5.1e-8
2020-10-15	45.9441	45.9441	2.1e-8
2020-10-16	48.6567	48.6567	1.5e-8
2020-10-17	36.2008	36.2008	7.4e-8
2020-10-18	28.9196	28.9196	1.6e-7
2020-10-19	21.3587	21.3587	1.9e-7
2020-10-20	24.0607	24.0607	1.5e-7
2020-10-21	34.9928	34.9928	5.1e-8
2020-10-22	35.0006	35.0006	8.2e-8
2020-10-23	49.8569	49.8569	2.4e-8
2020-10-24	41.0719	41.0719	6.4e-8
2020-10-25	28.9196	28.9196	1.6e-7
2020-10-26	21.3587	21.3587	1.9e-7
2020-10-27	35.0006	35.0006	8.2e-8
2020-10-28	34.9928	34.9928	5.1e-8
2020-10-29	47.1442	47.1442	2.9e-8
2020-10-30	48.6567	48.6567	1.5e-8
2020-10-31	41.0719	41.0719	6.4e-8
2020-11-01	35.0006	35.0006	8.2e-8
2020-11-02	21.3587	21.3587	1.9e-7
2020-11-03	24.0607	24.0607	1.5e-7
2020-11-04	36.1930	36.1930	4.3e-8
2020-11-05	45.9441	45.9441	2.1e-8
2020-11-06	35.0006	35.0006	8.2e-8
2020-11-07	41.0719	41.0719	6.4e-8
2020-11-08	28.9196	28.9196	1.6e-7
2020-11-09	21.3587	21.3587	1.9e-7
2020-11-10	25.2609	25.2609	1.4e-7
2020-11-11	35.0006	35.0006	8.2e-8
2020-11-12	45.9441	45.9441	2.1e-8
2020-11-13	48.6567	48.6567	1.5e-8
2020-11-14	41.0719	41.0719	6.4e-8
2020-11-15	28.9196	28.9196	1.6e-7

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Fit	3.82 ms	8.99 ms	2.4× faster
Predict	476 µs	4.71 ms	9.9× faster
Save model to JSON	18 µs	822 µs	45× faster
Load model from JSON	176 µs	3.84 ms	22× faster
Load and predict in a new process	158 ms	1.40 s	8.8× faster

Settings

The case declaration both libraries ran.

{
  "kind": "map",
  "comparison": {
    "kind": "equivalent-objective",
    "evidenceId": "flat-growth-mixed-scaling-v1"
  },
  "configuration": {
    "growth": "flat",
    "scaling": "absmax",
    "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
  }
}