matbench_rf.py

import os
import numpy as np
import warnings
import json

from time import time
from datetime import datetime

from sklearn.ensemble import RandomForestRegressor

from crabnet.utils.get_core_count import get_core_count
from crabnet.utils.modelselectionhelper import modelselectionhelper
from crabnet.utils.utils import NumpyEncoder, CONSTANTS, count_gs_param_combinations


# %%
cons = CONSTANTS()
mat_props_dir = r"data/matbench_cv/"
mat_props = cons.matbench_props

mat_props_names = cons.matbench_names
mat_props_pretty = cons.matbench_names_dict
elem_props = cons.eps
elem_props = ["magpie"]

RNG_SEED = 42
np.random.seed(RNG_SEED)


# %%
models1 = {
    "RandomForestRegressor": RandomForestRegressor(),
}

rf_estimators_range = [500]

params1 = {
    "RandomForestRegressor": {"n_estimators": rf_estimators_range, "max_depth": [None]},
}

scorings = {
    "r2": "r2",
    "neg_MAE": "neg_mean_absolute_error",
    "neg_RMSE": "neg_root_mean_squared_error",
}


# %%
if __name__ == "__main__":
    start_datetime_matbench_classics = datetime.now().strftime("%Y-%m-%d-%H%M%S.%f")

    metrics_dir = f"metrics/rf_gridsearch/"
    fig_dir = r"figures/GridSearchCV/matbench/"
    os.makedirs(metrics_dir, exist_ok=True)
    os.makedirs(fig_dir, exist_ok=True)

    _, cnt_dict = count_gs_param_combinations(params1)
    print(f"Number of parameter combinations for each estimator:\n" f"{cnt_dict}")

    cnt_dict_filename = "parameter_combos_matbench.json"
    cnt_dict_file = os.path.join(metrics_dir, cnt_dict_filename)
    json_content = json.dumps(cnt_dict, cls=NumpyEncoder, indent=4)

    with open(cnt_dict_file, "w") as f:
        try:
            f.write(json_content)
        except:
            pass

    n_cores = get_core_count()
    n_jobs = n_cores // 2 - 2

    ignore_warnings = True
    if ignore_warnings:
        maxiter_msg = (
            "Maximum number of iteration reached "
            "before convergence. Consider increasing max_iter "
            "to improve the fit."
        )
        warnings.filterwarnings("ignore", message=maxiter_msg)

    ti_matbench_classics = time()

    cv_folds = 2
    mshelper1 = modelselectionhelper(
        models1,
        params1,
        elem_props,
        mat_props_dir,
        mat_props,
        metrics_dir,
        fig_dir,
        scoring=scorings,
        n_jobs=n_jobs,
        cv=cv_folds,
        refit="neg_MAE",
        verbose=True,
        random_seed=RNG_SEED,
    )
    dt_matbench_classics = time() - ti_matbench_classics

    print("*********** matbench_classics finished ***********")
    print(
        f"matbench_classics finished, elapsed time: " f"{dt_matbench_classics:0.4g} s"
    )
    print("*********** matbench_classics finished ***********")