def objective(args):
parser = ArgumentParser()
parser.add_argument("-i",
"--input_file",
type=str,
required=True,
help="Path to data file")
parser.add_argument("-o",
"--output_dir",
type=str,
required=True,
help="Directory where model will be saved")
parser.add_argument("-r",
"--result_folder",
type=str,
default="./result/",
required=False,
help="Folder of result")
parser.add_argument("-d_e",
"--drug_encoding",
type=str,
default="Transformer",
required=False,
help="Drug encoding")
tmp = parser.parse_args()
base_config = vars(tmp) #把参数转换为字典格式
params = {
"input_file":
base_config["input_file"],
"output_dir":
base_config["output_dir"],
"result_folder":
base_config["result_folder"],
"drug_encoding":
base_config["drug_encoding"],
"input_dim_drug":
args["input_dim_drug"],
"input_dim_protein":
args["input_dim_protein"],
"hidden_dim_drug":
args["hidden_dim_drug"],
"hidden_dim_protein":
args["hidden_dim_protein"],
"cls_hidden_dims": [1024, 1024, 512],
"batch_size":
args["batch_size"],
"train_epoch":
10,
"test_every_X_epoch":
20,
"LR":
0.0001,
"decay":
0,
"num_workers":
0,
"transformer_dropout_rate":
args["transformer_dropout_rate"],
"transformer_emb_size_drug":
args["transformer_emb_size_drug"],
"transformer_n_layer_drug":
args["transformer_n_layer_drug"],
"transformer_intermediate_size_drug":
args["transformer_intermediate_size_drug"],
"transformer_num_attention_heads_drug":
args["transformer_num_attention_heads_drug"],
"transformer_attention_probs_dropout":
args["transformer_attention_probs_dropout"],
"transformer_hidden_dropout_rate":
args["transformer_hidden_dropout_rate"],
"transformer_emb_size_target":
args["transformer_emb_size_target"],
"transformer_n_layer_target":
args["transformer_n_layer_target"],
"transformer_intermediate_size_target":
args["transformer_intermediate_size_target"],
"transformer_num_attention_heads_target":
args["transformer_num_attention_heads_target"]
}
test_result = train_model(params)
return 1 - test_result["AUROC"]
# config of drug
parser.add_argument("--mlp_hidden_dims_drug", type = list, default = [1024, 256, 64], required = False,\
help = "Dimensionality of hidden layers of drug in MLP")
return parser
#模型配置生成
def get_model_config(config):
model_config = generate_config(
drug_encoding=config.drug_encoding,
result_folder=config.result_folder,
input_dim_drug=config.input_dim_drug,
hidden_dim_drug=config.hidden_dim_drug,
cls_hidden_dims=config.cls_hidden_dims,
batch_size=config.batch_size,
train_epoch=config.train_epoch,
test_every_X_epoch=config.test_every_X_epoch,
LR=config.LR,
decay=config.decay,
num_workers=config.num_workers,
mlp_hidden_dims_drug=config.mlp_hidden_dims_drug)
return model_config
if __name__ == "__main__":
parser = get_parser()
config = parser.parse_known_args()[0]
train_model(config)
hp.choice("hidden_dim_protein", [128, 256, 512]),
"batch_size":
hp.choice("batch_size", [32, 64, 128, 256, 512]),
"transformer_dropout_rate":
hp.choice("transformer_dropout_rate", [0.1]),
"transformer_emb_size_drug":
hp.choice("transformer_emb_size_drug", [128, 256]),
"transformer_n_layer_drug":
hp.choice("transformer_n_layer_drug", range(1, 10)),
"transformer_intermediate_size_drug":
hp.choice("transformer_intermediate_size_drug", [128, 256, 512]),
"transformer_num_attention_heads_drug":
hp.choice("transformer_num_attention_heads_drug", [2, 4, 8]),
"transformer_attention_probs_dropout":
hp.choice("transformer_attention_probs_dropout", [0.1]),
"transformer_hidden_dropout_rate":
hp.choice("transformer_hidden_dropout_rate", [0.1]),
"transformer_emb_size_target":
hp.choice("transformer_emb_size_target", [32, 64, 128]),
"transformer_n_layer_target":
hp.choice("transformer_n_layer_target", range(1, 10)),
"transformer_intermediate_size_target":
hp.choice("transformer_intermediate_size_target", [128, 256, 512]),
"transformer_num_attention_heads_target":
hp.choice("transformer_num_attention_heads_target", [2, 4, 8])
}
best = fmin(fn=objective, space=space, algo=tpe.suggest, max_evals=100)
print(best)
train_model(best)