def test_save_model():
dset = get_test_dataset()
args = {'featureflag': 'FCHL',
'targetflag': 'HCS'}
params = {'cutoff': 5.0}
dset.get_features_frommols(args, params)
model = FCHLmodel()
model.get_x(dset, args['targetflag'], assign_train=True)
model.train()
filename='tests/test_tmp/tmp_model.pkl'
model.save_model(filename=filename)
assert os.path.isfile(filename)
ld_model = FCHLmodel()
ld_model.load_model(filename)
train_x, train_y = model.get_x(dset, args['targetflag'], assign_train=False)
test_x = np.asarray(train_x[:, :4])
y_pred1 = ld_model.predict(test_x)
y_pred2 = model.predict(test_x)
assert np.array_equal(y_pred1, y_pred2)
os.remove(filename)
def test_trainFCHLmodel():
dset = get_test_dataset()
for target in ['HCS', '1JCH']:
args = {'featureflag': 'FCHL',
'targetflag': target}
params = {'cutoff': 5.0}
dset.get_features_frommols(args, params)
model = FCHLmodel()
model.get_x(dset, args['targetflag'], assign_train=True)
model.train()
assert model.trained == True
def test_iteration():
dataset = get_test_dataset()
args = {
'featureflag': 'CMAT',
'targetflag': 'HCS',
'feature_optimisation': 'True',
'logfile': 'tests/test_tmp/tmp_logfile',
'searchflag': 'test_search',
'modelflag': 'KRR',
'param_list': ['cutoff', 'sigma', 'lamda'],
'param_logs': {
'cutoff': 'lin',
'sigma': 'log',
'lamda': 'log'
},
'param_ranges': {
'cutoff': [0, 10],
'sigma': [-5, -2],
'lamda': [-10, -4]
},
'output_dir': './'
}
params = {'cutoff': 5.0, 'sigma': 0.1, 'lamda': 0.01}
model = KRRmodel()
BEST_SCORE = 99999.9999
BEST_PARAMS = {}
HPS_generic.setup_logfile(args)
for iter in range(2):
next_point_to_probe = params
score, BEST_SCORE, BEST_PARAMS = HPS_generic.HPS_iteration(
model,
iter,
dataset,
args,
next_point_to_probe=next_point_to_probe,
BEST_SCORE=BEST_SCORE,
BEST_PARAMS=BEST_PARAMS)
print(score, BEST_SCORE, BEST_PARAMS)
os.remove(args['logfile'])
def test_random_hyperopt():
dataset = get_test_dataset()
args = {'featureflag': 'CMAT',
'targetflag': 'HCS',
'feature_optimisation': 'True',
'logfile': 'tests/test_tmp/tmp_logfile',
'searchflag': 'test_search',
'modelflag': 'KRR',
'param_list': ['cutoff', 'sigma', 'lamda'],
'param_logs': {'cutoff': 'lin', 'sigma': 'log', 'lamda': 'log'},
'param_ranges': {'cutoff': [0, 10], 'sigma': [-5, -2], 'lamda': [-10, -4]},
'output_dir': './'}
args['epochs'] = 3
model = KRRmodel()
dataset, BEST_SCORE = HPS_random.random_search(model, dataset, args)
def test_cvpredict_FCHLmodel_shift():
dset = get_test_dataset(size=6)
for target in ['HCS', '1JCH']:
args = {'featureflag': 'FCHL',
'targetflag': target}
params = {'cutoff': 5.0}
dset.get_features_frommols(args, params)
model = KRRmodel()
model.get_x(dset, args['targetflag'], assign_train=True)
model.train()
assert model.trained == True
y_pred = model.cv_predict(fold=3)
assert y_pred.shape == model.train_y.shape
def test_predictFCHLmodel():
dset = get_test_dataset()
for target in ['HCS', '1JCH']:
args = {'featureflag': 'FCHL',
'targetflag': target}
params = {'cutoff': 5.0}
dset.get_features_frommols(args, params)
model = FCHLmodel()
target = flag_to_target(args['targetflag'])
model.get_x(dset, args['targetflag'], assign_train=True)
model.train()
assert model.trained == True
train_x, train_y = model.get_x(dset, args['targetflag'], assign_train=False)
test_x = np.asarray(train_x[:, :4])
y_pred = model.predict(test_x)
assert np.allclose(y_pred, train_y[:4], atol=0.1, rtol=0.0)
assert sum(y_pred) != 0.0
