This is an implementation of a DYNAMIS-like approach to dynamic problem solving. It creates an instance of a Problem which is a linear structural equation. A problem has some number of input nodes, some number of output nodes, and a set of relationships between them. Given control of the inputs, and the values of the outputs, can you determine the relationships between them?
When creating a new problem the following can be set to tune the difficulty.
inputNotesThe number of user controlled inputsoutputNodesThe number of outputs that will be exposed.hiddenNodesThe number of hidden variables in the network. Beware: these are insidious.relationshipsHow many relationships should exist in the network. Input nodes never have ingress relationships.allowSelfReferenceDefault = true. Can a node have a relationship with itself?
Creating a new problem with 2 inputs, 2 outputs, and 4 relationship edges. Relationships may have the same node for source and target. No hidden nodes.
curl -vX POST http://localhost:8080/problems -d @request.json --header "Content-Type: application/json"
request.json
{
"inputNodes" : 2,
"outputNodes" : 2,
"relationships" : 4,
"allowSelfReference" : true
}
Get the graphviz DOT file for a given problem (essentially the solution).
curl http://localhost:8080/problems/{id}/dot
Get the current data points for a problem.
curl http://localhost:8080/problems/{id}
Advance the current problem one step without modifying inputs.
curl -vX POST http://localhost:8080/problems/{id}/next --header "Content-Type: application/json"
Advance the current problem with specified inputs. Inputs are all named Input_ with zero-indexed names. Any unnamed inputs will be left at their previous value. If a variable isn't found it will raise an exception.
curl -vX POST http://localhost:8080/problems/{id}/next -d @values.json --header "Content-Type: application/json"
values.json
{
"Input_0": 0.0
}