add launch from python code doc #3217

J-shang · 2020-12-22T01:46:06Z

TODO: modify experiment config url

kvartet · 2020-12-30T11:59:06Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+Overview
+--------
+This is a new way to launch experiments since ``nni v2.0``. Before, you need to configure our experiment in a yaml configuration file and launch the experiment through ``nnictl`` command. Now, you can also configure and run your experiment directly in the python file. If you are familiar with python programming, this will undoubtedly bring you more convenience.


our experiment -> your experiment?

kvartet · 2020-12-30T12:01:18Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+    experiment = Experiment(tuner=foo_tuner, training_service='local')
+
+Now, you have a ``Experiment`` instance with ``foo_tuner`` you have initialized in the previous step, and this experiment will launch trials on your local machine due to ``training_service='local'``.


I saw training_service. Name of training service. Supported value: "local", "remote", "openpai" from the code comments, I don't know whether it supports other training services.

kvartet · 2020-12-30T12:04:33Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+Example
+-------
+Below is an example for this new launching approach. You can also find this code in ``examples/trials/mnist-tfv2/launch.py``.


maybe

:githublink:`mnist-tfv2/launch.py <examples/trials/mnist-tfv2/launch.py>`

will be better.

kvartet · 2020-12-30T12:09:05Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+    experiment.config.training_service.use_active_gpu = True
+
+    experiment.run(8081)
+


Perhaps we should add a note to indicate whether experiments launched from python can be managed by nnictl or nni client.

liuzhe-lz · 2020-12-31T07:18:05Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+..
+
+    Step 1 - Initialize a tuner you want to use


I think "step by step" is not the best way to introduce a library, seems too verbose...

I think it is a quick start to a new way to launch experiment, not just a library, like https://nni.readthedocs.io/en/v2.0/Tutorial/QuickStart.html#three-steps-to-start-an-experiment

Well I think writing Python program and using a "ctl" tool is different. But I'm okay with current version.

QuanluZhang · 2021-01-02T14:44:05Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+Overview
+--------
+This is a new way to launch experiments since ``nni v2.0``. Before, you need to configure your experiment in a yaml configuration file and launch the experiment through ``nnictl`` command. Now, you can also configure and run your experiment directly in the python file. If you are familiar with python programming, this will undoubtedly bring you more convenience.


"the python file" -> "a python file"

QuanluZhang · 2021-01-02T14:45:53Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+.. code-block:: python
+
+    from nni.algorithms.hpo.foo_tuner import FooTuner


do we really have FooTuner? better to use a real and simple tuner as example

QuanluZhang · 2021-01-02T14:46:34Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+..
+
+    Step 2 - Initialize a experiment instance and configure it


"a experiment" -> "an experiment"

QuanluZhang · 2021-01-02T14:49:17Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+.. code-block:: python
+
+    experiment.config.experiment_name = 'test'


better to make this part more real, for example, setting up a real local experiment

QuanluZhang · 2021-01-02T14:49:30Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+
+Use the form like ``experiment.config.foo = 'bar'`` to configure your experiment.
+
+See `parameter configuration <TBD>`__ required by different training services.


here is a broken link

waiting for #3222

QuanluZhang · 2021-01-02T14:51:21Z

docs/en_US/Tutorial/HowToLaunchFromPython.rst

+..  autoclass:: nni.experiment.Experiment
+    :members:
+
+..  autoclass:: nni.experiment.TrialJob


can this class be used in the new launch approach?

I will check it

QuanluZhang · 2021-01-02T14:51:58Z

nni/experiment/experiment.py

@@ -69,7 +69,7 @@ def __init__(self, tuner: Tuner, training_service: str) -> None:
            A tuner instance.
        training_service
            Name of training service.
-            Supported value: "local", "remote", "openpai".
+            Supported value: "local", "remote", "openpai", "aml", "kubeflow", "frameworkcontroller", "adl".


so we support all the training services in v2.0?

It seems we don't support heterogeneous and DLTS? @liuzhe-lz

you can add "and hybrid training service."

scarlett2018

I made some change suggestions directly to the Howto.rst file.

add launch from python code

0df10a3

liuzhe-lz requested review from scarlett2018 and liuzhe-lz December 23, 2020 02:45

scarlett2018 requested a review from kvartet December 25, 2020 01:34

liuzhe-lz requested a review from QuanluZhang December 30, 2020 03:01

scarlett2018 mentioned this pull request Dec 30, 2020

v2.0 Release Plan #2935

Closed

77 tasks

kvartet reviewed Dec 30, 2020

View reviewed changes

J-shang added 2 commits December 31, 2020 12:36

update doc

300cea9

move ref in nnicli to launch from python

ab50126

liuzhe-lz reviewed Dec 31, 2020

View reviewed changes

liuzhe-lz approved these changes Dec 31, 2020

View reviewed changes

QuanluZhang reviewed Jan 2, 2021

View reviewed changes

J-shang added 4 commits January 3, 2021 22:12

fix typo

e2f2dd2

add exp config link

1776113

Merge branch 'doc-launch' of github.com:J-shang/nni into doc-launch

65b5943

update API

261c842

QuanluZhang approved these changes Jan 4, 2021

View reviewed changes

Update HowToLaunchFromPython.rst

3a82abb

scarlett2018 approved these changes Jan 5, 2021

View reviewed changes

fix lint

9c7066f

QuanluZhang merged commit b4f0d32 into microsoft:v2.0 Jan 5, 2021

J-shang deleted the doc-launch branch January 7, 2021 05:50

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add launch from python code doc #3217

add launch from python code doc #3217

J-shang commented Dec 22, 2020 •

edited

Loading

kvartet Dec 30, 2020

J-shang Dec 31, 2020

kvartet Dec 30, 2020 •

edited

Loading

J-shang Dec 31, 2020

kvartet Dec 30, 2020

J-shang Dec 31, 2020

kvartet Dec 30, 2020 •

edited

Loading

liuzhe-lz Dec 31, 2020

J-shang Dec 31, 2020

liuzhe-lz Dec 31, 2020

QuanluZhang Jan 2, 2021

J-shang Jan 3, 2021

QuanluZhang Jan 2, 2021

J-shang Jan 3, 2021

QuanluZhang Jan 2, 2021

J-shang Jan 3, 2021

QuanluZhang Jan 2, 2021

J-shang Jan 3, 2021

QuanluZhang Jan 2, 2021

J-shang Jan 3, 2021

QuanluZhang Jan 2, 2021

J-shang Jan 3, 2021

QuanluZhang Jan 2, 2021

J-shang Jan 3, 2021

QuanluZhang Jan 4, 2021

J-shang Jan 5, 2021

scarlett2018 left a comment


		experiment = Experiment(tuner=foo_tuner, training_service='local')

		Now, you have a ``Experiment`` instance with ``foo_tuner`` you have initialized in the previous step, and this experiment will launch trials on your local machine due to ``training_service='local'``.

		experiment.config.training_service.use_active_gpu = True

		experiment.run(8081)


		.. code-block:: python

		from nni.algorithms.hpo.foo_tuner import FooTuner


		..

		Step 2 - Initialize a experiment instance and configure it


		.. code-block:: python

		experiment.config.experiment_name = 'test'


		Use the form like ``experiment.config.foo = 'bar'`` to configure your experiment.

		See `parameter configuration <TBD>`__ required by different training services.

add launch from python code doc #3217

add launch from python code doc #3217

Conversation

J-shang commented Dec 22, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kvartet Dec 30, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kvartet Dec 30, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

scarlett2018 left a comment

Choose a reason for hiding this comment

J-shang commented Dec 22, 2020 •

edited

Loading

kvartet Dec 30, 2020 •

edited

Loading

kvartet Dec 30, 2020 •

edited

Loading