saquaia python interface

0) installing saquaia

Prerequisites: java, maven. Install saquaia by running

cd saquaia-mvn-main/code
mvn package

Verify installation by running

java -jar target/saquaia-jar-with-dependencies.jar

Script saqint.py demonstrates the usage of saquaia from Python (see main())

python3 saqint.py

1) Models

From the 4 models available at Nessie, only ts and mapk are supported by saquaia. main() contains their corresponding text descriptions, which can be parsed into our internal representation for future conversion. For each model, one must specify the initial state (see Nessie) and species bounds (if unknown, just use some large number).

2) Creating benchmarks

BenchmarkFactory allows you to create multiple instantiations of the same benchmark template, e.g. trying different parameter valuation, different number of simulations, different methods (vanilla SSA vs our SEG), seed for saquaia simulations, c parameter for SEG (worth playing with: values above 1 are accepted, smaller = more precise), etc. prefix will be used as a prefix for benchmark names. BenchmarkFactory::new will create a specific benchmark given a specific method name and parameter valuation; the suffix is appended to the prefix to obtain the name of the benchmark. Each benchmark is basically a json object compatible with saquaia interface.

3) Evaluating benchmarks

saquaia_run method will run saquaia (transient analysis) on a given benchmark and return probability distribution over states and runtime (s). Under the hood, I store the benchmark object to a (temporary) file, run saquaia from the command line, saquaia generates a ton of outputs to saquaia-result/BENCHMARK_NAME folder, from which I parse the results. Since I call saquaia from the command line, running saqint.py from another location might break saquaia since it uses relative names for output: let me know if you need this fixed. By default, all the temporary files are removed unless cleanup=False is passed to saquaia_run.