Usage

The command line, seamm-al, is described in the user guide. The library can also be used directly from Python, for example to score structures with a committee:

from ase.io import read
from active_learning_step.committee import (
    Committee, masses, molecules, structure_from_atoms
)

committee = Committee.from_checkpoints(["m0/best.pt", "m1/best.pt", "m2/best.pt"])
frames = [structure_from_atoms(a) for a in read("cells.extxyz", index=":")]
spreads = committee.score(frames, batch_size=1, molecule=molecules, masses=masses)
print(spreads["rho_forces"], spreads["rho_pressure_molecular"])

and to choose frames from scores:

from active_learning_step.selection import choose, parse_costs, read_scores

rows = read_scores("scores.csv")
selected, rejected = choose(rows, parse_costs(["cell=2800"]), budget=75000)