Test-case for asymmetric FCHL

test_asymmetric.py

from __future__ import print_function

import os

import ast
import csv

from scipy.linalg import lstsq

import qml
from qml.fchl import generate_representation
from qml.fchl import get_atomic_local_kernels

np.random.seed(667)

CSV_FILE = "force_test.csv"

TRAINING = 100
TEST     = 101

CUT_DISTANCE = 1e6
SIGMAS = [2.5] 
KERNEL_ARGS = {
    "verbose": False,
    "cut_distance": CUT_DISTANCE,
    "kernel": "gaussian",
    "kernel_args": {
            "sigma": SIGMAS,
    },      
}   


def mae(a, b):

    return np.mean(np.abs(a.flatten() - b.flatten()))


def csv_to_reps(csv_filename):

    x = []
    e = []

    with open(csv_filename, 'r') as csvfile:

        df = csv.reader(csvfile, delimiter=";", quotechar='#')

        for row in df:
    
            coordinates = np.array(ast.literal_eval(row[3]))
            nuclear_charges = ast.literal_eval(row[8])
            energy = float(row[1])

            rep = generate_representation(coordinates, nuclear_charges, cut_distance=CUT_DISTANCE)

            x.append(rep)
            e.append(energy)
            
    return np.array(x), np.array(e)


def test_asymmetric():

    Xall, Eall, = csv_to_reps(CSV_FILE)

    X  = Xall[:TRAINING]
    Xs = Xall[-TEST:]

    E  = Eall[:TRAINING]
    Es = Eall[-TEST:]

    K  = get_atomic_local_kernels(X, X,   **KERNEL_ARGS)
    Ks = get_atomic_local_kernels(X, Xs,  **KERNEL_ARGS)
    
    for i, sigma in enumerate(SIGMAS):
    
        alphas, residuals, singular_values, rank = lstsq(K[i].T, E, cond=1e-9, lapack_driver="gelsd")

        # Test energy prediction
        Ess = np.dot(Ks[i].T, alphas)

        # Training energy predictions
        Et  = np.dot(K[i].T,  alphas)

        print(mae(Ess, Es))
        print(mae(Et, E))


if __name__ == "__main__":

    test_asymmetric()