Capacity, floor and a multiplicative regressor

match ← All logistic growth benchmarks

96 daily rows with changing capacities, an explicit floor, min-max scaling, weekly seasonality, a multiplicative regressor and two automatic changepoints. mcmc_samples=300.

Input

benchmark/inputs/v1/logistic-explicit-floor.json: 96 history rows and 24 forecast rows.

Output

Largest difference: 0.0044 (0.0091% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-04-06	65.8320	65.8337	0.0017
2020-04-07	66.2473	66.2481	0.00078
2020-04-08	67.2872	67.2830	0.0042
2020-04-09	67.5700	67.5714	0.0014
2020-04-10	67.9574	67.9587	0.0013
2020-04-11	68.0044	68.0037	0.00072
2020-04-12	67.8983	67.8983	5.8e-5
2020-04-13	67.9422	67.9440	0.0018
2020-04-14	68.6967	68.6934	0.0033
2020-04-15	68.9348	68.9347	8.2e-5
2020-04-16	69.5493	69.5508	0.0015
2020-04-17	69.8956	69.8969	0.0014
2020-04-18	69.9027	69.9020	0.00069
2020-04-19	69.7578	69.7578	4.4e-5
2020-04-20	70.1568	70.1544	0.0024
2020-04-21	70.0982	70.0991	0.00084
2020-04-22	70.6851	70.6850	9.2e-5
2020-04-23	71.2655	71.2669	0.0014
2020-04-24	71.5787	71.5801	0.0013
2020-04-25	71.5539	71.5531	0.00072
2020-04-26	71.7785	71.7741	0.0044
2020-04-27	71.3547	71.3564	0.0018
2020-04-28	71.6599	71.6607	0.00080
2020-04-29	72.2189	72.2188	0.00014

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Fit	1.83 s	452 ms	4.0× slower
Predict	396 µs	3.79 ms	Not compared: different work
Save model to JSON	269 µs	12.5 ms	46× faster
Load model from JSON	2.87 ms	8.01 ms	2.8× faster
Simulate uncertainty	2.91 ms	210 ms	Not compared: different work
Load and simulate in a new process	186 ms	1.65 s	Not compared: different work

Some steps aren't compared, because the two libraries' public calls produce different outputs. Their times show what each call costs, not which library is faster.

• Predict: Python's predict on a posterior also computes every component's range over the draws, while Effect Prophet's predict returns point forecasts and leaves ranges to predictWithIntervals.
• Simulate uncertainty, Load and simulate in a new process: Python's predict(vectorized=False) also builds the point forecast and every component, while Effect Prophet's predictUncertainty returns only the ranges.

Settings

The case declaration both libraries ran.

{
  "kind": "map",
  "comparison": {
    "kind": "equivalent-posterior",
    "evidenceId": "posterior-nuts-v1"
  },
  "configuration": {
    "growth": "logistic",
    "scaling": "minmax",
    "seasonalityMode": "additive",
    "holidaysMode": "additive",
    "changepoints": {
      "mode": "auto",
      "count": 2,
      "range": 0.8
    },
    "changepointPriorScale": 0.2,
    "seasonalities": [
      {
        "name": "weekly-custom",
        "periodDays": 7,
        "fourierOrder": 2,
        "priorScale": 10
      }
    ],
    "events": [],
    "regressors": [
      {
        "name": "promotion",
        "priorScale": 10,
        "standardization": "never",
        "mode": "multiplicative"
      }
    ]
  },
  "pythonOptimizer": {
    "algorithm": "Auto",
    "maxIterations": 10000,
    "newtonFallback": true,
    "sigFigs": 12
  },
  "uncertainty": {
    "seed": 19,
    "samples": 1000,
    "intervalWidth": 0.8,
    "output": "intervals",
    "algorithm": "scalar-continuous-time",
    "evidence": "docs/validation/uncertainty.md#evidence-classes",
    "samplerMemoryLimitBytes": 67108864
  },
  "mcmc": {
    "samples": 300,
    "seed": 19,
    "pointWidthFraction": 0.15
  }
}