(1) Fix /0 bug in double_ensemble, (2) remove _default_uri for R/expm, (3) support model size in pytorch models by D-X-Y · Pull Request #314 · microsoft/qlib

D-X-Y · 2021-03-08T06:27:57Z

Four changes:

Fix /0 bug in double_ensemble
simplify the logic of default_uri for R/expm
support model size in pytorch models

Description

In double_ensemble models, it might potentially divide zero and raise an error.
Currently, _default_uri for R maybe inconsistent with that in C.
In pytorch-based contrib codes, there are some useless kwargs, such as verbose.
In pytorch-based contrib codes, add a function to count the number of parameters.

Motivation and Context

Make it more friendly for researchers.

How Has This Been Tested?

Pass the test by running: pytest qlib/tests/test_all_pipeline.py under upper directory of qlib.

Screenshots of Test Results (if appropriate):

I run pytest -W ignore::DeprecationWarning qlib/tests/test_all_pipeline.py, below is the screenshot:

Types of changes

Fix bugs
Add new feature

Derek-Wds · 2021-03-08T07:21:15Z

qlib/workflow/expm.py

        )

+    def reset_default_uri(self, uri: Text):
+        self._default_uri = uri


Is any reason that you want to add this method? From my perspective, default_uri should be initialized and fixed when users call qlib.init. It seems there is no need to have such method here.

A simple case is:
after we run many different experiments (with different uri), we may want to compare their performance. To query the metrics of each of these experiments, we need to let R know their uri.

@D-X-Y We have uri in C, default_uri and current_uri here.
I think this will make users confusing.
Do you think combining uri in C and default_uri is a good idea?

Derek-Wds · 2021-03-08T07:23:50Z

qlib/workflow/recorder.py

+        return self._artifact_uri
+
+    @property
+    def root_uri(self):


What is this method for? Can you give us more explanation about it?

This is to return the path of a specific record. For example, you can call R.get_recorder().root_uri and it will return: ml-runs/2/32792162f0f44951b1285c0b2e489252.

Please update the related docs.

Derek-Wds · 2021-03-08T07:24:57Z

qlib/workflow/__init__.py

        """
        return self.exp_manager.uri

+    def reset_default_uri(self, uri: Text):


Same as the one in expm.

Yeap, to provide an api to reset default uri. Use to query experiments with different uri.

qlib/contrib/model/pytorch_utils.py

Derek-Wds · 2021-03-08T07:43:48Z

qlib/log.py

    logging_config.dictConfig(log_config)


+def set_log_basic_config(filename: Optional[Text] = None, format: Optional[Text] = None, level: Optional[int] = None):


Most of the loggers used in Qlib are initialized through the function get_module_logger. If you want to do some enhancement about setting the logger, you may want to make some modifications upon that function instead of creating a new one. This function seems to be a little bit redundant.

The motivation is that get_module_logger is called in different separate Python files, and it is hard for users to control the detailed behavior of get_module_logger function in each file.
This function provides a global control for logging, a case is that before each experiment runs, we can use this function to save the following logs into a file under the recorder directory (without caring about the kwargs of get_module_logger in each file).

We think config logging using Config will make it easier for user to manage all the configurations.
https://github.com/microsoft/qlib/blob/main/qlib/config.py#L266

@D-X-Y
You can set it in the C instead of create a new function

We think the original Python logging system is not user-friendly.
Qlib may use loguru in the future #7

you-n-g · 2021-03-08T08:31:02Z

qlib/contrib/model/pytorch_utils.py

+import torch.nn as nn
+
+
+def count_parameters(models_or_parameters, unit="mb"):


What do you mean by mb?
If it is a storage size unit. the byte length of each number should be considered.
Otherwise, please give explanation of the b

If it is a storage size unit, please use 2**10 instead of 1e3

you-n-g · 2021-03-08T08:42:10Z

qlib/log.py

    logging_config.dictConfig(log_config)


+def set_log_basic_config(filename: Optional[Text] = None, format: Optional[Text] = None, level: Optional[int] = None):


We think config logging using Config will make it easier for user to manage all the configurations.
https://github.com/microsoft/qlib/blob/main/qlib/config.py#L266

@D-X-Y
You can set it in the C instead of create a new function

you-n-g · 2021-03-08T08:43:24Z

qlib/log.py

    logging_config.dictConfig(log_config)


+def set_log_basic_config(filename: Optional[Text] = None, format: Optional[Text] = None, level: Optional[int] = None):


We think the original Python logging system is not user-friendly.
Qlib may use loguru in the future #7

you-n-g · 2021-03-08T08:58:48Z

qlib/workflow/expm.py

        )

+    def reset_default_uri(self, uri: Text):
+        self._default_uri = uri


@D-X-Y We have uri in C, default_uri and current_uri here.
I think this will make users confusing.
Do you think combining uri in C and default_uri is a good idea?

you-n-g · 2021-03-08T09:03:57Z

qlib/workflow/recorder.py

+    def root_uri(self):
+        start_str = "file:"
+        if self.artifact_uri is not None:
+            xpath = self.artifact_uri.strip(start_str)


lstrip will be more accurate

you-n-g · 2021-03-08T09:07:19Z

qlib/workflow/recorder.py

+        return self._artifact_uri
+
+    @property
+    def root_uri(self):


If the funtion aims to get the path in a filesystem, it should raise error(e.g. NotImplementError) when the uri is not file system.

Besides, the root_uri is not a good name for user.

D-X-Y · 2021-03-09T09:14:54Z

@you-n-g do you have any suggested name for root_uri?

… as get_local_dir

D-X-Y · 2021-03-11T03:12:33Z

Following our offline discussion, I have:

temporarily removed set_log functions
simplify the logic of default uri in expm
rename root_uri to get_local_dir with more robust logic and doc
update count_parameters and add doc

Please let me know if you have any other concerns about the current version.

qlib/contrib/backtest/account.py

Derek-Wds

LGTM. Thanks~

D-X-Y · 2021-03-11T06:47:30Z

@Derek-Wds Thanks for your prompt response.

…icrosoft#314)

(1) Fix /0 bug in double_ensemble, (2) remove _default_uri for R/expm, (3) support model size in pytorch models

D-X-Y added 11 commits March 5, 2021 07:15

Add set_log_basic_config function support re-directing log stream

19d9374

Merge branch 'fshare' of github.com:D-X-Y/qlib into fshare

4f980a0

Merge branch 'main' of github.com:microsoft/qlib into fshare

b40bfb1

Merge branch 'main' of github.com:D-X-Y/qlib into fshare

b14a559

Add count_parameters for pytorch models in contrib

131f0e2

Show model size for pytorch models

49697b1

Add reset_default_uri func for R and expm

91fd53a

Remove useless verbose kwarg

73b7107

Avoid dividing zero in model.double_ensemble

d13c9ae

Merge branch 'main' of github.com:microsoft/qlib into fshare

4266492

Fix python format by black

7bed3b4

Derek-Wds added bug Something isn't working enhancement New feature or request labels Mar 8, 2021

Derek-Wds reviewed Mar 8, 2021

View reviewed changes

D-X-Y added 3 commits March 8, 2021 08:16

Simplify count_parameters

ca48345

Fix bugs in count_parameters

03ef918

Fix lint error with Black

e061443

you-n-g reviewed Mar 8, 2021

View reviewed changes

D-X-Y added 2 commits March 11, 2021 02:33

Remove set_log_basic_config, refine count_parameters, rename root_uri…

f6ed175

… as get_local_dir

Refine default uri in expm

cda96be

Fix bugs for get_local_dir

0ef7c8e

D-X-Y requested review from Derek-Wds and you-n-g March 11, 2021 03:10

Derek-Wds reviewed Mar 11, 2021

View reviewed changes

qlib/contrib/backtest/account.py Show resolved Hide resolved

D-X-Y requested a review from Derek-Wds March 11, 2021 04:54

Derek-Wds approved these changes Mar 11, 2021

View reviewed changes

you-n-g merged commit e8beaa5 into microsoft:main Mar 11, 2021

you-n-g pushed a commit to you-n-g/qlib that referenced this pull request Aug 16, 2025

align local data folder and kaggle data folder to store mid result (m…

f6c2f77

…icrosoft#314)

gaogen123 pushed a commit to gaogen123/qlib that referenced this pull request Sep 13, 2025

Merge pull request microsoft#314 from D-X-Y/fshare

f605b34

(1) Fix /0 bug in double_ensemble, (2) remove _default_uri for R/expm, (3) support model size in pytorch models

		logging_config.dictConfig(log_config)


		def set_log_basic_config(filename: Optional[Text] = None, format: Optional[Text] = None, level: Optional[int] = None):

		import torch.nn as nn


		def count_parameters(models_or_parameters, unit="mb"):

Conversation

D-X-Y commented Mar 8, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Motivation and Context

How Has This Been Tested?

Screenshots of Test Results (if appropriate):

Types of changes

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

D-X-Y commented Mar 9, 2021

Uh oh!

D-X-Y commented Mar 11, 2021

Uh oh!

Uh oh!

Derek-Wds left a comment

Choose a reason for hiding this comment

Uh oh!

D-X-Y commented Mar 11, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

D-X-Y commented Mar 8, 2021 •

edited

Loading