| title | description | services | documentationcenter | author | manager | editor | ms.assetid | ms.service | ms.workload | ms.tgt_pltfrm | ms.devlang | ms.topic | ms.date | ms.author |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Cortana Intelligence Gallery experiments | Microsoft Docs |
Discover and share experiments in the Cortana Intelligence Gallery. |
machine-learning |
garyericson |
jhubbard |
cgronlun |
f4248922-c961-4d3a-9e1b-aec743210166 |
machine-learning |
data-services |
na |
na |
article |
10/13/2016 |
roopalik;garye |
[!INCLUDE machine-learning-gallery-item-selector]
The Gallery contains a wide variety of Experiments that have been developed in Azure Machine Learning Studio. These range from quick proof-of-concept experiments that demonstrate a specific machine learning technique, to fully-developed solutions for complex machine learning problems.
Note
An experiment is a canvas in Azure Machine Learning Studio that lets you construct a predictive analysis model by connecting together data with various analytical modules. You can try different ideas, do trial runs, and eventually publish your model as a web service in Azure. For an example of creating a simple experiment, see Machine learning tutorial: Create your first experiment in Azure Machine Learning Studio. For a more complete walkthrough of creating a predictive analytics solution, see Walkthrough: Develop a predictive analytics solution for credit risk assessment in Azure Machine Learning.
To browse for experiments in the Gallery, open the Gallery and click Experiments at the top of the Gallery home page.
The Experiments page displays a list of the most recently added and most popular experiments. Click See all to view all experiments. From this page you can browse all the experiments in the Gallery, or you can search by selecting filter criteria on the left of the page and entering search terms at the top.
Click any experiment to open the experiment's details page and read information about what the experiment does. On this page you can comment, provide feedback, or ask questions through the comments section. You can even share it with friends or colleagues using the share capabilities of LinkedIn or Twitter. You can also email a link to the experiment to invite other users to view the page.
You can download a copy of any experiment from the Gallery into your Studio workspace and then modify your copy to create your own solutions. There are two ways to get a copy of the experiment:
- From the Gallery - If you find an experiment you like in the Gallery, you can easily download a copy and open it in your Machine Learning Studio workspace.
- From within Machine Learning Studio - In Studio, you can use any experiment in the Gallery as a template to create a new experiment.
To download a copy of an experiment from the Gallery:
-
Open the experiment's details page in the Gallery
-
Click Open in Studio
When you click Open in Studio, the experiment is loaded into your Machine Learning Studio workspace and opened (if you're not already signed in to Studio, you will be prompted to sign in using your Microsoft account before the experiment is copied to your workspace).
You can also open the same sample experiments while you're working in Machine Learning Studio:
-
In Machine Learning Studio, click +NEW
-
Select Experiment - you can choose from a list of Gallery experiments contributed by Microsoft, or you can find a specific experiment using the search box
-
Point your mouse at an experiment and select Open in Studio - the experiment is copied to your workspace and opened (to see information about the experiment, select View in Gallery which takes you to the details page for the experiment in the Gallery)
You can now customize, iterate, and deploy this experiment like any other experiment you create in Machine Learning Studio.
When you sign in to the Gallery you become a member of the Gallery community. This allows you to contribute your own experiments so that others can benefit from the solutions you've discovered.
Follow these steps to contribute an experiment to the Cortana Intelligence Gallery:
-
Sign in to Machine Learning Studio using your Microsoft account.
-
Create your experiment and run it.
-
When you’re ready to publish your experiment to the Gallery, click Publish to Gallery below the experiment canvas.
-
Fill out the title and tags fields. Keep them descriptive, highlighting the techniques used or the real-world problem being solved, for instance, “Binary Classification: Twitter Sentiment Analysis”.
-
Write a summary of what your content covers. Briefly describe the problem being solved and how you approached it.
-
Use the detailed description box to step through the different parts of your experiment. Some useful topics to include here are:
-
Experiment graph screenshot
-
Data sources and explanation
-
Data processing
-
Feature engineering
-
Model description
-
Results and evaluation of model performance
You can use Markdown to format as needed. Click the Preview icon to see how things will look when published.
[!TIP] The box provided for Markdown editing and preview box is quite small. We recommend that you write your documentation in a Markdown editor and paste the completed document into the text box. After you've published your experiment, you can use standard web-based tools in Markdown for editing and preview to make necessary tweaks and corrections.
-
-
Upload a thumbnail image for your gallery item. This will appear at the top of the item page and in the item tile when browsing the gallery. You can choose an image from your computer or select one of the stock images.
-
Choose whether to publish your content publicly, or have it only accessible to people with the link.
[!TIP] If you want to make sure your documentation looks right before releasing it publicly, you can publish it as unlisted first, and then switch it to Public from the item page.
-
Click the OK checkmark to publish the experiment to the Gallery.
See the section below, Suggestions for publishing and for quality documentation, for tips on how to publish a quality Gallery experiment.
That’s it – you’re all done.
You can now view your experiment in the Gallery and share the link with others. If you published it publicly, your experiment will show up in browse and search results in the Gallery. You can also edit your documentation on the item page any time you're logged in.
To see the list of your contributions, click your image in the upper-right corner of any Gallery page and then click your name to open your account page.
If you need to make changes to the workflow (modules, parameters, etc.) in an experiment you published to the Gallery, go back to the experiment in Machine Learning Studio, make your changes, and publish it again. Your existing published experiment will be updated with your changes.
If you just need to change any of the following information for your experiment, or you need to delete the experiment from the Gallery, you can make all of these changes in the Gallery:
- Experiment name
- Summary or description text
- Tags specified
- Image used
- Visibility setting (public or unlisted)
- Delete experiment from the Gallery
These changes can be made in the Gallery from the experiment's details page or from your profile page.
From the experiment's details page, click "edit" to change the details for your experiment.
The details page enters edit mode, and you can click "Edit" next to the experiment name, summary, tags, etc., to make changes to them. When you've finished making changes, click "Done".
You can also select the settings icon to change visibility of the experiment (public or unlisted) or you can delete the experiment from the Gallery.
From your profile page, you can click the down arrow on the experiment and select "Edit". This takes you to the details page for your experiment in edit mode. When you have finished making changes, click "Done".
You can also click "Delete" to delete the experiment from the Gallery.
- While you can assume that the reader has prior data science experience, it still helps to simplify your language and explain things in detail wherever possible.
- Not all readers will be familiar with the Cortana Intelligence Suite, given that it's relatively new. So provide enough information and step-by-step explanations to help readers navigate through your work.
- Visuals including experiment graphs or screenshots of data can be very helpful for readers to interpret and use your content the right way. See the Publishing Guidelines and Examples collection for more information on how to include images in your documentation.
- If you include a dataset in your experiment (it's not being imported through the Import Data module), it's part of your experiment and will get published to the Gallery. Therefore, ensure that the dataset you're publishing has appropriate licensing terms for sharing and downloading by anyone. Gallery contributions are covered under the Azure Terms of Use.
What are the image requirements when submitting or editing an image for my experiment?
The images you submit along with your experiment will be used to create an experiment tile for your contribution. It's recommended that the images be < 500Kb in size, with an aspect ratio of 3:2. A resolution of 960x640 is recommended
What happens to the dataset I used in the experiment? Does the dataset get published to the Gallery as well?
If your dataset is part of your experiment and not being imported through the Import Data module, it's part of your experiment and gets published to the Gallery with your experiment. For this reason make sure that the dataset you're publishing with the experiment has the appropriate licensing terms that allow sharing and downloading by anyone.
I have an experiment that uses an Import Data module to pull data from HDInsight or SQL. It uses my credentials to retrieve the data. How can I publish such an experiment and be assured that my credentials will not be shared?
At this time we do not allow publishing of experiments that use credentials.
How do I enter multiple tags?
After you enter a tag, press the tab key to enter another tag.
[!INCLUDE machine-learning-free-trial]