def train_all(train_dataset, vali_dataset):
train(DecisionTreeClassifierTrainer(), train_dataset, vali_dataset)
train(DecisionTreeTrainer(), train_dataset, vali_dataset)
train(LassoTrainer(), train_dataset, vali_dataset)
train(LogisticRegressionTrainer(), train_dataset, vali_dataset)
train(SGDClassifierTrainer(), train_dataset, vali_dataset)
if "vali_set" not in params:
logging.error('"vali_set" not defined')
exit(1)
if "test_set" not in params:
logging.error('"test_set" not defined')
exit(1)
# device = torch.device("cuda") if torch.cuda.is_available() else None
device = None
num_columns = params["num_columns"] if "num_columns" in params else 138
anchor_url_features = (params["anchor_url_features"]
if "anchor_url_features" in params else None)
body_features = params[
"body_features"] if "body_features" in params else None
train_dataset = load_dataset(params["train_set"], num_columns,
anchor_url_features, body_features)
vali_dataset = load_dataset(params["vali_set"], num_columns,
anchor_url_features, body_features)
train_all(train_dataset, vali_dataset)
exit(0)
test_dataset = load_dataset(params["test_set"], num_columns,
anchor_url_features, body_features)
evalute_all(test_dataset, 5, DecisionTreeTrainer(), DecisionTreeTrainer(),
True, 100, 100)
def train_all(train_dataset, vali_dataset, prefix: str = ""):
# train(DecisionTreeClassifierTrainer(), train_dataset, vali_dataset)
train(DecisionTreeTrainer(), train_dataset, vali_dataset, prefix)
train(LassoTrainer(), train_dataset, vali_dataset, prefix)
description="Read command line parameters.")
parser.add_argument("-p", "--parameters", help="Path to config file.")
args = parser.parse_args(sys.argv[1:])
with open(args.parameters, "r") as f:
params = json.load(f)
if "dataset" not in params:
raise Exception('Please define "dataset" in config file')
random.seed(1234)
np.random.seed(1234)
torch.random.manual_seed(1234)
dataset = UCIMultiClassDataset(params["dataset"])
log_trainer = LogisticRegressionTrainer()
log_epsilon = 0.1
tgt_trainer = SGDClassifierTrainer()
tgt_epsilon = 0.1
dm_trainer = DecisionTreeTrainer()
experiments = [(
(
DMEstimator(DecisionTreeTrainer()),
IPSEstimator(),
DoublyRobustEstimator(DecisionTreeTrainer()),
),
1000,
) for _ in range(100)]
evaluate_all(experiments, dataset, log_trainer, log_epsilon, tgt_trainer,
tgt_epsilon, 0)
body_features = params[
"body_features"] if "body_features" in params else None
# uncomment to train behavior and target models
# train_models(params)
test_dataset = load_dataset(
params["second_set"],
num_columns,
anchor_url_features,
body_features,
"second_set",
)
weight_clamper = Clamper(min_v=0.0)
estimators = [
DMEstimator(DecisionTreeTrainer(), 0.5, device=device),
IPSEstimator(weight_clamper=weight_clamper, device=device),
DoublyRobustEstimator(DecisionTreeTrainer(), 0.5, weight_clamper,
False, device),
DoublyRobustEstimator(DecisionTreeTrainer(), 0.5, weight_clamper, True,
device),
PseudoInverseEstimator(weight_clamper=weight_clamper, device=device),
PBMEstimator(weight_clamper=weight_clamper, device=device),
]
metrics = ["ndcg", "err"]
alphas = [0.0, 1.0, 2.0]
trainers = [
(DecisionTreeTrainer(), LassoTrainer()),
(LassoTrainer(), DecisionTreeTrainer()),
]