]Analysis: 1. level
COCO-OPTI2 online - standard version
Preconditions for using COCO-OPTI2:
- There is a given OAM.
- During its development, the determination of vectors were inevitable.
- The number of stairs is no more than 10 % of the objects the minimum however is 5 (because from the model certain restrictive conditions are going to be omitted in order to break the monotone stair-figures at one/multiple points). Too many stairs may not necessarily lead to advantageous polynomials. In case of too low number of stairs the optimum-effect will be too rough.
- The Y values are integer numbers, (with or without shift) and are non-negative.
- The Y do not have to be zero if any of the attributes are missing (=additive effect).
- Columns that have the same effect were connected in order to speed up the execution (cf. repetitive columns lose their effect during the execution).
- The first rows of stairs might mean the model's (genetic) potential.
- Ceteris paribus functions can be non-monotone. (Basically they have one extreme value as maximum.)
- MCM or stair function without restriction of the stairs: If we specify an OAM with relatively low number of stairs, and we happen to be curious about, what kind of ceteris paribus figures can be imagined to be in the background of a certain learning pattern, then it is unnecessary to assess how shall a better place be related with a worse one. In case of 3 stairs all stair executions are interpretable. If there are more, stairs may wave (cf. polynomial-effect). If the number of stairs is equal with the number of objects, then we get a classic MCM, (or Monte-Carlo Method / basically unguided search control) that try to grope the combinatorial space with quasi random numbers (further details and the demo of the LP-precipitation with partially goal-oriented search control.
- It is worth taking into consideration in case of optimizing models that with decision tree-like classifications the robustness (cf. COCO STEP) of the solution can be enhanced behind the seemingly confused ceteris paribus fluctuations.
Areas of application:
- Building models that allows presuming optimum-effects (e.g. production functions)
- Simulation of decision trees (viz. searching for polynomial-parts that can be explained with the IF/THEN principle)
- Revelation of system-behavior
- Making forecasts.
- Not advised: in case of benchmarking and price/performance examination!
