liuyu/qlib
1
0
Fork 0
mirror of https://github.com/microsoft/qlib.git synced 2026-07-11 14:56:55 +08:00
Code Activity

the second version of online serving

Browse Source
This commit is contained in:
lzh222333
2021-03-12 08:04:08 +00:00
parent 0df88c07f6
commit 6d8aa215d6
5 changed files with 75 additions and 127 deletions
Download Patch File Download Diff File Expand all files Collapse all files

90
qlib/workflow/task/collect.py
View File

@@ -2,6 +2,7 @@ from qlib.workflow import R
import pandas as pd
from typing import Union
from typing import Callable
from qlib import get_module_logger
@@ -17,13 +18,13 @@ class TaskCollector:
def list_recorders(self, rec_filter_func=None, task_filter_func=None, only_finished=True, only_have_task=False):
"""
Return a dict of {rid:recorder} by recorder filter and task filter. It is not necessary to use those filter.
If you don't train with "task_train", then there is no "task.pkl" which includes the task config.
If there is a "task.pkl", then it will become rec.task which can be get simply.
Return a dict of {rid:Recorder} by recorder filter and task filter. It is not necessary to use those filter.
If you don't train with "task_train", then there is no "task" which includes the task config.
If there is a "task", then it will become rec.task which can be get simply.
Parameters
----------
rec_filter_func : Callable[[MLflowRecorder], bool], optional
rec_filter_func : Callable[[Recorder], bool], optional
judge whether you need this recorder, by default None
task_filter_func : Callable[[dict], bool], optional
judge whether you need this task, by default None
@@ -35,30 +36,27 @@ class TaskCollector:
Returns
-------
dict
a dict of {rid:recorder}
a dict of {rid:Recorder}
Raises
------
OSError
if you use a task filter, but there is no "task.pkl" which includes the task config
if you use a task filter, but there is no "task" which includes the task config
"""
recs = self.exp.list_recorders()
# return all recorders if the filter is None and you don't need task
if rec_filter_func==None and task_filter_func==None and only_have_task==False:
return recs
recs_flt = {}
if task_filter_func is not None:
only_have_task = True
for rid, rec in recs.items():
if (only_finished and rec.status == rec.STATUS_FI) or only_finished==False:
if rec_filter_func is None or rec_filter_func(rec):
task = None
try:
task = rec.load_object("task.pkl")
task = rec.load_object("task")
except OSError:
if task_filter_func is not None:
raise OSError('Can not find "task.pkl" in your records, have you train with "task_train" method in qlib.model.trainer?')
pass
if task is None and only_have_task:
continue
if task_filter_func is None or task_filter_func(task):
rec.task = task
recs_flt[rid] = rec
@@ -68,7 +66,7 @@ class TaskCollector:
def collect_predictions(
self,
get_key_func,
filter_func=None,
task_filter_func=None,
):
"""
@@ -85,7 +83,7 @@ class TaskCollector:
dict
the dict of predictions
"""
recs_flt = self.list_recorders(task_filter_func=filter_func)
recs_flt = self.list_recorders(task_filter_func=task_filter_func,only_have_task=True)
# group
recs_group = {}
@@ -108,11 +106,14 @@ class TaskCollector:
def collect_latest_records(
self,
filter_func=None,
task_filter_func=None,
):
recs_flt = self.list_recorders(task_filter_func=filter_func,only_have_task=True)
max_test = max(rec.task['dataset']['kwargs']['segments']['test'] for rec in recs_flt.values())
recs_flt = self.list_recorders(task_filter_func=task_filter_func,only_have_task=True)
if len(recs_flt) == 0:
self.logger.warning("Can not collect any recorders...")
return None, None
max_test = max(rec.task['dataset']['kwargs']['segments']['test'] for rec in recs_flt.values())
latest_record = {}
for rid, rec in recs_flt.items():
@@ -120,52 +121,5 @@ class TaskCollector:
latest_record[rid] = rec
self.logger.info(f"Collect {len(latest_record)} latest records in {self.exp_name}")
return latest_record
class RollingCollector:
"""
Rolling Models Ensemble based on (R)ecord
This shares nothing with Ensemble
"""
# TODO: speed up this class
def __init__(self, get_key_func, flt_func=None):
self.get_key_func = get_key_func # get the key of a task based on task config
self.flt_func = flt_func # determine whether a task can be retained based on task config
def __call__(self, exp_name) -> Union[pd.Series, dict]:
# TODO;
# Should we split the scripts into several sub functions?
exp = R.get_exp(experiment_name=exp_name)
# filter records
recs = exp.list_recorders()
recs_flt = {}
for rid, rec in tqdm(recs.items(), desc="Loading data"):
params = rec.load_object("task.pkl")
if rec.status == rec.STATUS_FI:
if self.flt_func is None or self.flt_func(params):
rec.params = params
recs_flt[rid] = rec
# group
recs_group = {}
for _, rec in recs_flt.items():
params = rec.params
group_key = self.get_key_func(params)
recs_group.setdefault(group_key, []).append(rec)
# reduce group
reduce_group = {}
for k, rec_l in recs_group.items():
pred_l = []
for rec in rec_l:
pred_l.append(rec.load_object("pred.pkl").iloc[:, 0])
pred = pd.concat(pred_l).sort_index()
reduce_group[k] = pred
return reduce_group
return latest_record, max_test