Describe the issue or request
A clear and concise description or request for enhancement.

Additional context
Add any other context about the problem here.

What notebook?
Add training example using a FileDataset and a SRC

Additional context
Add any other context here.

What notebook?
What would you like to see in an example notebook?

Additional context
Add any other context here.

Some Tutorial notebooks do not have any explanation but only have codes. We need to set up a bar for a notebook to be a part of Tutorials.

in order to qualify as tutorial, notebook should have:

requirements (what you need to understand / run)
"deliverables" (what you get from running the tutorial)
some kind of written content (not code) talking through what's happening

Some examples:

• https://github.com/Azure/azureml-examples/blob/main/tutorials/using-dask/2.eds-at-scale.ipynb
• https://github.com/Azure/azureml-examples/blob/main/notebooks/train-lightgbm-local.ipynb

What notebook?
refactor for easier automation - look into RG delete/create?

Additional context
Add any other context here.

Describe the issue or request
for:

• issue templates
• other?

Additional context
Add any other context about the problem here.

The script setup.py logs error messages

Here are some examples:

Failure while loading azureml_run_type_providers. Failed to load entrypoint hyperdrive = azureml.train.hyperdrive:HyperDriveRun._from_run_dto with exception (azureml-telemetry 1.17.0 (/home/amin/miniconda/lib/python3.7/site-packages), Requirement.parse('azureml-telemetry~=1.15.0')).

Failure while loading azureml_run_type_providers. Failed to load entrypoint automl = azureml.train.automl.run:AutoMLRun._from_run_dto with exception (azureml-telemetry 1.17.0 (/home/amin/miniconda/lib/python3.7/site-packages), Requirement.parse('azureml-telemetry~=1.15.0')).

Guessing these are caused by our requirements.txt file not "pinning" dependency versions, since this doesn't reproduce in a fresh Conda environment.

Should we "nudge" users towards a clean Conda or virtual environment? In addition, maybe "pin" package versions? This will be especially relevant as SDK VNext lands.

Especially relevant since our notebooks / code don't mention version information.

Could use "dependabot" to grab new Python package versions.

Likely related to: Task 907711: AutoML: Pin version of the portalocker package to 1.7.0.

The script setup.py does not have enough documentation

It "does things", people will want to understand more details. It doesn't need to be much, but high-level.

This repo does a really nice job with "Getting started" and "Contents":

https://github.com/microsoft/aml-acceleration-template

e.g. clear instructions on how to use an existing workspace.

Describe the issue or request
if pull_request is added to the triggers, it fails because forked repos don't have the Azure credentials. Need to investigate how to work around this.

Additional context

Describe the issue or request
replace samples AML experience with this and other examples

Additional context
Add any other context about the problem here.

What example? Describe it

example: lightgbm gpu
description: single node single gpu, single node multi gpu, multi node multi gpu

Additional context

• using python api

Which example?
using-rapids multi GPU tutorial notebook

Describe the issue
scheduled testing is failing

To Reproduce
see tests: https://github.com/Azure/azureml-examples/actions?query=workflow%3Arun-tutorial-ur

Expected behavior
test passes

Screenshots

Additional context
Add any other context about the problem here.

Describe the issue or request
we are supporting a wide range of scenarios and, particularly for GPUs, need different compute targets

add to setup.ipynb, configure in testing workspace

Additional context
Add any other context about the problem here.

Which notebook?
fastai/train-pets-resnet34.ipynb

Describe the issue
not working

To Reproduce
run the notebook

Expected behavior
notebook runs

Screenshots
If applicable, add screenshots to help explain your problem.

Environment details

• any

Additional context
Add any other context about the problem here.

When users come across a repo with setup.py in the root, they expect to be able to "install the repo" via pip install .. However, that is not what this repository's setup.py does.

Recommendation: name it something like setup-aml-infrastructure.py.

What notebook?
Add pipeline example(s)

Additional context
Add any other context here.

Describe the issue or request
With links to new notebooks

Additional context
Add any other context about the problem here.

Currently, setup.py has a hardcoded default location of eastus2 for the workspace. It should have some basic logic to default to the user's location (if possible, not sure this is).

In theory, this could be done calling ARM client SDK - but's that's super slow. Hoping there's a fast way.

Another way to handle this would be to make setup.py interactive (could use cookie cutter, or have some config file for this).

Best - use Hydra for configuring setup.py. Cleanly isolates what setup.py does from how it does it.

• ensure readme.py is modified
• use upload artifact github action

@lostmygithubaccount are we planning to add a config notebook that will create the compute targets referenced in the examples?

Which example? Describe the issue

example: several
description: submitting run, not checking if it completes successfully or fails

Additional context

• need to add to contributing guidelines

Describe the issue

When I follow along the using-dask tutorial to setup a Dask cluster, I fail to create a new cluster with an odd error that comes up:
[Errno 113] No route to host

I tried searching for it, but the results on Google will come with port and routing issues, that I don't know if are directly related to the problem here. Not sure.

Minimal example

from azureml.core import Workspace

ws = Workspace.from_config()

from azureml.core import Environment

env = Environment.get(workspace=ws, name="AzureML-Dask-CPU")

from dask.distributed import Client
from dask_cloudprovider import AzureMLCluster

cluster = AzureMLCluster(
    ws,
    vm_size="STANDARD_DS5_V2",
    environment_definition=env,
    initial_node_count=10,
    scheduler_idle_timeout=1200,
)

c = Client(cluster)
c

Additional context

• Important to note that I am running all of this inside an Azure ML Compute Instance

Which example?
rapids hyperdrive

Describe the issue
runs get stuck - need to set max experiment timeout

To Reproduce
Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

Expected behavior
A clear and concise description of what you expected to happen.

Screenshots
If applicable, add screenshots to help explain your problem.

Environment details

• OS: [e.g. Windows 10, Ubuntu]
• Azure ML SDK Version [e.g. 1.10.0]
• other relevant details

Additional context
Add any other context about the problem here.

Describe the issue or request
Add a PR template

Additional context
none

Describe the issue or request
Add testing for all current Python versions

Additional context
need to consider resources needed for this

Which notebook?
a few of them

Describe the issue
NN is in the description when a CNN is trained

To Reproduce
read

Which example? Describe the issue

example: tutorials/using-dask
description: fails in automated testing

Additional context

• see failures

cleanup:

• workspaces (from concept)
• compute targets
• compute instances
• webservices

What notebook?
What would you like to see in an example notebook?

Additional context
Add any other context here.

Describe the issue or request
the az ml installation command used in this repo (mostly in workflows) is not the publicly documented method

Additional context
this issue needs to be resolved before switching back: Azure/azure-cli#15434

Which notebook?
notebooks/triton/deploy densenet local.ipynb

Describe the issue
azureml-core is pinned to 1.15.0 via pip install in the notebook

To Reproduce
Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

Expected behavior
uses latest SDK installed from requirements.txt

Screenshots
If applicable, add screenshots to help explain your problem.

Environment details

• OS: [e.g. Windows 10, Ubuntu]
• Azure ML SDK Version [e.g. 1.10.0]
• other relevant details

Additional context
Add any other context about the problem here.

What notebook?
What would you like to see in an example notebook?
Showcase pytorch DDP distributed training

Additional context
Add any other context here.

Describe the issue or request
A clear and concise description or request for enhancement.

Additional context
Add any other context about the problem here.

What notebook?
Add distributed training tensorflow example

Additional context
Add any other context here.

There's plenty of existing material to draw from.

Describe the issue or request
switch to using uuid or something more fancy instead of randint -> str and appending it

Additional context
Add any other context about the problem here.

What example? Describe it

example: xgboost gpu examples
description: single node single gpu, single node multi gpu, multi node multi gpu

Additional context

• using python api

Which example? Describe the issue

example: using-rapids tutorial
description: test failing

Additional context

• need to investigate

Which example?
Which example is the issue in?

Describe the issue
A clear and concise description of what the issue is.

To Reproduce
Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

Expected behavior
A clear and concise description of what you expected to happen.

Screenshots
If applicable, add screenshots to help explain your problem.

Environment details

• OS: [e.g. Windows 10, Ubuntu]
• Azure ML SDK Version [e.g. 1.10.0]
• other relevant details

Additional context
Add any other context about the problem here.

Describe the issue or request
A clear and concise description or request for enhancement.

Additional context
Add any other context about the problem here.

Which notebook?
concepts/datastore

Describe the issue
uses TuringNLR datastore

To Reproduce
read

Expected behavior

Screenshots

Environment details

• OS: [e.g. Windows 10, Ubuntu]
• Azure ML SDK Version [e.g. 1.10.0]
• other relevant details

Additional context

Which example? Describe the issue

example: an-introduction/1.hello-world.ipynb
description: under understanding the code there are 2 minor issues

Additional context

• link to Run Python API docs broken
• experiment -> exp per code cell

Describe the issue or request
A clear and concise description or request for enhancement.

Additional context
Add any other context about the problem here.

The example train/deepspeed-cifar10.py requires compute target gpu-V100-2

This seems like overkill and makes the examples less accessible. Is it really necessary?

More generally, most subscriptions start out with a V100 quota of 0. The process for increasing this is painful and lengthy. Most all examples in the repo don't really need such heavy compute.

Also: consider provisioning low-pri compute (at least: mention).

What notebook?
AutoML thing client

Additional context
Add any other context here.

Describe the issue or request
A clear and concise description or request for enhancement.

Additional context
Add any other context about the problem here.

Describe the issue or request
cheat sheet pivot stays highlighted - i suspect moving into its own folder may fix?

Additional context
see:

Hopefully title is self-descriptive.

Even if users technically have quota, they'll be worried about cost. Also, nothing is gained.

Suggest: 4 node max. (Can put comment saying to put more if wanted.)