Rombus is a tool for building reduced order models (ROMs): matrix representations of arbitrary models which can be rapidly and easily computed for arbitrary parameter sets.
Building a ROM with Rombus is easy. All you need to do is install it like so:
$ pip install rombus
define your model like this (in this trivial case, a file named my_model.py
specifying a simple second-order polynomial):
from numpy import ndarray, polyval, linspace
from rombus.model import RombusModel
from typing import NamedTuple
class Model(RombusModel):
"""Class for creating a ROM for the function y(x)=a2*x^2+a1*x+a0"""
coordinate.set("x", 0.0, 10.0, 11, label="$x$")
ordinate.set("y", label="$y(x)$")
params.add("a0", -10, 10)
params.add("a1", -10, 10)
params.add("a2", -10, 10)
def compute(self, p: NamedTuple, x: ndarray) -> ndarray:
"""Compute the model for a given parameter set."""
return polyval([p.a2, p.a1, p.a0], x)
and specify a set of points (in this case, the file my_model_samples.py
) to build your ROM from:
-10, -10,-10
-10, 10,-10
-10, -10, 10
-10, 10, 10
10, -10,-10
10, 10,-10
10, -10, 10
10, 10, 10
You build your ROM like this:
$ rombus build my_model:Model my_model_samples.csv
This produces an HDF5 file named my_model.hdf5
. You can then use your new ROM in your Python projects like this:
from rombus.rom import ReducedOrderModel
ROM = ReducedOrderModel.from_file('my_model.hdf5')
sample = ROM.model.sample({"a0":0,"a1":0,"a2":1})
model_ROM = ROM.evaluate(sample)
for x, y in zip(ROM.model.domain,model_ROM):
print(f"{x:5.2f} {y:6.2f}")
which generates the output:
0.00 0.00
1.00 1.00
2.00 4.00
3.00 9.00
4.00 16.00
5.00 25.00
6.00 36.00
7.00 49.00
8.00 64.00
9.00 81.00
10.00 100.00