def test_parse_grid(self):
config1 = parse_params(GOOD_YAML_FILE, "grid")
config2 = {
"max_depth": [3, None],
"max_features": [1, 11],
"min_samples_split": [2, 11],
"bootstrap": [0, 1],
"criterion": ["gini", "entropy"],
"learning_rate": [1e-06, 10.0]
}
assert config1 == config2
def test_parse_bayes(self):
config1 = parse_params(GOOD_YAML_FILE, "bayes")
config2 = {
"max_depth": Categorical([3, None]),
"max_features": Integer(1, 11),
"learning_rate": Real(1e-06, 10.0)
}
assert type(config1["max_depth"]) == type(config2["max_depth"])
assert isinstance(config1["learning_rate"],
type(config2["learning_rate"]))
assert isinstance(config1["max_features"],
type(config2["max_features"]))
def test_parse_randomized(self):
config1 = parse_params(GOOD_YAML_FILE, "randomized")
config2 = {
"max_depth": [3, None],
"max_features": randint(1, 11),
"learning_rate": uniform(1e-06, 10.0)
}
assert isinstance(config1["max_depth"], type(config2["max_depth"]))
assert isinstance(config1["learning_rate"],
type(config2["learning_rate"]))
assert isinstance(config1["max_features"],
type(config2["max_features"]))
def test_badly_formatted_yaml(self):
with pytest.raises(ParserError):
parse_params(BAD_YAML_FILE, "grid")
def test_parse_with_unkown_mathod(self):
with pytest.raises(ValueError):
parse_params(GOOD_YAML_FILE, "foobar")
parser.add_argument("--scale",
type=int,
help="Number of cluster workers to requesat.")
return parser.parse_args()
if __name__ == '__main__':
args = parse_args()
# replace with your data
digits = load_digits()
X, y = digits.data, digits.target
# replace with your model
model = RandomForestClassifier(n_estimators=20)
# replace with your hyperparameter search config
config = os.path.join(CURRENT_DIR, "params.yaml")
params = parse_params(config, args.method)
# set up context manager for distributing computations
LOGGER = logging.getLogger(__name__)
CLUSTER = launch_cluster(args.cluster, args.scale)
CLIENT = Client(CLUSTER)
LOGGER.warn("Web dashboard now running at http://localhost:8787/status")
with joblib.parallel_backend('dask.distributed'):
result = search(model, X, y, params, method=args.method)
# try to get progress bar in terminal; prefer web daskboad
progress(result)
dump(result, args.filepath)