Skip to content

Latest commit

 

History

History
81 lines (59 loc) · 3.27 KB

README.md

File metadata and controls

81 lines (59 loc) · 3.27 KB
Raw

Master Thesis

Evolutionary Computation in Hierarchical Model Discovery

David Revillas, 2022

Requirements

Requires Python 3.8+. Install dependencies (virtual environment recommended) with:

pip install -U -r requirements.txt

If pip command is linked to a Python 2 environment, try:

python3 -m pip install -U -r requirements.txt

Execution

usage: benchmark.py [-h] --bench {temperature,drivers,precipitation,normal,beta} [--day {0,1,2,3,4,5,6,7,8,9}] [--generations GENERATIONS] [--id ID] [-r REPEAT] [--no-figures]
                    [-l {simple,inference}] [-o {svi,botorch}] [--continue-from CONTINUE_FROM]

Benchmark

optional arguments:
  -h, --help            show this help message and exit
  --bench {temperature,drivers,precipitation,normal,beta}
                        'temperature': Launches the `2.1 Average Minimum Temperature in Scotland` problem benchmark.
                        'precipitation': Models the precipitation in Punta Galea
  --generations GENERATIONS, -g GENERATIONS
                        Number of generations to evolve (default: 100)
  --id ID               Folder name to store results (default: current datetime)
  -r REPEAT, --repeat REPEAT
                        Repeats n times the experiment (default: 1)
  --no-figures          Doesn't save figures
  -l {simple,inference}, --loss {simple,inference}
                        Select the loss function:
                        'simple' performs parameter inference only on the best evolved individual.
                        'inference', inference is applied to every single individual in the evolution (slower). (Default: simple)
  -o {svi,botorch}, --optimizer {svi,botorch}
                        Select the input optimizer: 'svi' performs SVI posterior distribution on the best evolved individual.
                        'botorch', numerical optimization (not implemented yet) (Default: None)
  --continue-from CONTINUE_FROM
                        Continues experiment number n (default: 0)

Executed experiments

Here are shown the shell commands executed for the experiments shown in the memory.

Problem 2.1

Before proceeding with these experiments, it is important to exclude Normal and Beta distributions respectively from the primitive set before launching them.

# Evolves during 250 generations the 'Normal' experiment with 'simple' loss function, repeating 5 times the entire process
python src/benchmark.py --bench normal --repeat 5 --generations 250 --id svi --loss simple --optimizer svi

# Evolves during 250 generations the 'Beta' experiment with 'simple' loss function, repeating 5 times the entire process
python src/benchmark.py --bench beta --repeat 5 --generations 250 --id svi --loss simple --optimizer svi

Problem 2.2

# Evolves during 250 generations the 'Temperature' experiment with 'simple' loss function, repeating 5 times the entire process
python src/benchmark.py --bench temperature --repeat 5 --generations 250 --id svi --loss simple --optimizer svi

Problem 2.3

# Evolves during 250 generations the 'Precipitation' experiment with 'simple' loss function, repeating 5 times the entire process
python src / benchmark . py -- bench precipitation -- repeat 5 --generations 250 -- id svi -- loss simple -- optimizer svi