update trainer and README.md

qlib/model/trainer.py

@@ -0,0 +1,40 @@
+# 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()