Assuming I have
Formula A = √(EF+EP+√(EP-NF))
Formula B = (NP*(NF+EF))/√(EP+NP)
Each individual formula produces their outputs respectively.
What kind of machine learning model can I use to learn from these two formulas and produce may be different formulas from the variables or produce better performing result than individual result.
Or may be a model that can combine the two formulas to produce better performing result.
Convert the formulas into trees. Write a recursive function that can evaluate the formula for concrete variable assignments. Consider the trees as DNA strands for a genetic algorithm. You can define the crossover mutation by exchanging sub-trees of different trees. To determine the fitness of an individual, evaluate a function for a set of variable values and sum up the squares of differences to the desired function values.
Regards,
Joachim
Use the automata theory and the same stuff as used in compiler construction.
1-) Parse this string left to write and get tokens.
2-) Build a tree out of these token a.k.a. parse tree.
3-) Finally, write the back end. If you want some automated tool look for https://www.antlr.org/
Instead of converting into some other language, the visitor pattern can be used to build a transformed tree. You can use certain heuristics to define the "better" or you could actually use some optimization theory to rebuild the transformed tree.
Adekunle Ajibode Please search for things like symbolic regression machine learning etc. Please have a look at these ones: Preprint AI Feynman: a Physics-Inspired Method for Symbolic Regression
, Preprint AI Feynman 2.0: Pareto-optimal symbolic regression exploitin...
,Preprint Symbolic Pregression: Discovering Physical Laws from Raw Dis...
,Preprint Deep Learning for Symbolic Mathematics
and this amazing article for an overview: https://www.quantamagazine.org/symbolic-mathematics-finally-yields-to-neural-networks-20200520/
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning