mirror of
https://github.com/microsoft/qlib.git
synced 2026-07-04 11:30:57 +08:00
41 lines
1.3 KiB
Python
41 lines
1.3 KiB
Python
# Copyright (c) Microsoft Corporation.
|
|
# Licensed under the MIT License.
|
|
from qlib.utils import init_instance_by_config, flatten_dict
|
|
from qlib.workflow import R
|
|
from qlib.workflow.record_temp import SignalRecord
|
|
|
|
|
|
def task_train(config: dict, experiment_name):
|
|
"""
|
|
task based training
|
|
|
|
Parameters
|
|
----------
|
|
config : dict
|
|
A dict describing the training process
|
|
"""
|
|
|
|
# model initiaiton
|
|
model = init_instance_by_config(config.get("task")["model"])
|
|
dataset = init_instance_by_config(config.get("task")["dataset"])
|
|
|
|
# start exp
|
|
with R.start(experiment_name=experiment_name):
|
|
# train model
|
|
R.log_params(**flatten_dict(config.get("task")))
|
|
model.fit(dataset)
|
|
recorder = R.get_recorder()
|
|
|
|
# generate records: prediction, backtest, and analysis
|
|
for record in config.get("task")["record"]:
|
|
if record["class"] == SignalRecord.__name__:
|
|
srconf = {"model": model, "dataset": dataset, "recorder": recorder}
|
|
record["kwargs"].update(srconf)
|
|
sr = init_instance_by_config(record)
|
|
sr.generate()
|
|
else:
|
|
rconf = {"recorder": recorder}
|
|
record["kwargs"].update(rconf)
|
|
ar = init_instance_by_config(record)
|
|
ar.generate()
|