This video walks through the steps to building a model in One AI, along with some explanation of how One AI fits into the rest of One Model.

For those of you who prefer to follow the steps in creating your own model in One AI, we've got you covered.  What follows is more focused on the entire process of getting a basic model up and running and less on settings that Data Scientists might want to manipulate.  There are some links provided for articles that get into more detail on particular subjects and we will be adding more of these detailed articles in time.

Determine what it is you would you like to predict

Technically, you can attempt to predict anything that the data elements you have available in One Model support.  To help you narrow it down, the team at One Model can provide ideas and guidance here.  Here are some examples:

  • Whether someone will be promoted or terminated (the classic attrition risk prediction)

  • Whether an applicant will receive an interview or an offer

  • Quality of hire

Get the supporting data modeled in One Model if it isn't already

You likely have enough data to run a prediction but do you have the right things?

  • Flags - For predicting terminations, you need to know whether people who were in headcount ultimately terminated.  If you do not already have a dimension or a column in your employee table called "Is Terminated" or "Is Future Terminated", then you will need one created.  

  • Outside data elements - Data points such as commute time for employees are generally not standard but may be able to be added by the One Model team if you wish to consider them in your prediction.

  • Generally, the more features you can add the better, since One AI will automatically choose the best ones.

  • The team at One Model will assist you in loading and/or modeling the data you need in order to make a high quality prediction.

Build the data sets for your prediction

This is done by first creating a "destination" and then creating and adding the data sets to it.  You will need to add 3 data sets to your One AI destination and each data set must contain at least four columns.  All three should match in content and structure.  The following help document covers destination files in more detail:
One AI - Data Destination Files

  1. Create the destination.  In One Model, go to Data > Destinations, Add Data Destination, and select One AI.

  2. To add the first data set to the destination, select the + for that destination and select Query Source.
    Note: The other option here is Processing Script.  The processing Script option is available so that the One Model team can create additional transformations on your already modeled data specifically for this data destination.

  3. Name it and Save and Explore Query.  Generally you'll want to name it one of the sample dates you intend to include, for example "2019".

  4. Build a query.  See the image below for an example of how it might look.  The query will require at least four columns.  They are as follows:
    Dataset ID: This is a unique id per row in the file. For example, you are creating an attrition risk model, this would likely be the employee id.
    Sample Date: This field denotes the "as of" date for the records in the file. For example, you might provide data from the beginning of 2017, the beginning of 2018, and the beginning of 2019 to predict attrition risk for 2019.
    Classifier Target: This is the column with the value you are trying to predict. For example, it might have a 0 or a 1 to indicate whether the employee terminated. For a regression problem it might have a number in it, like headcount or days to fill. The format does not have to be numeric for classification problems. Instead of a 0 or 1, you might have a value that says "Terminated" or "Didn't Terminate".
    Features: Features (or attributes) are the things that influence the prediction.  Just having a list of 100 employees, 20 of which terminate does not indicate what caused those 20 to terminate.  We also need some defining attributes about the employees so we can try to determine what it is about these 20 that caused them to terminate.  You could add just one feature but generally the more the merrier since One AI will automatically choose the best ones for the prediction.

  5. Select the Pin icon above the results and Update data destination query.

  6. Now change your time period.  The case of this example I would change it to 2018 and Run the query.

  7. Select the Pin icon above the results and Add to data destination.  Select the data destination you created, name the file, and add it.  Again, generally you'll want to name it the time period of the query, for example "2018".

  8. Repeat steps 6 - 7 one more time, for your third file.  In this example it would be 2017.

  9. Run the destination.  To do this, navigate back to the Data Destinations screen and select the Run button next to the destination you created.

  10. Verify that the run completes by viewing data destination history, which is the button to the right of Run.

Here is an example of what a basic query might look like, although yours should contain many more feature columns:

Create the One AI model (Augmentation)

An augmentation is a configuration for the creation of a machine learning model.  One AI has the capability to determine the best possible configuration for many settings, but it needs a few basic items defined first.

  1. Navigate to Data > Augmentations and select Add Augmentation.

  2. Give it a name.

  3. Select the Augmentation Type you wish to create.  The augmentation types are as follows:
    Classification: Predicts whether or not something is likely to happen.  Will the employee terminate? Yes or No
    Regression: Predicts a count or value.  For example, the number of days until someone is likely to terminate.

  4. Select the data destination you just created.

  5. Note that the Refresh data from the selected destination button here will re-run your data destination.

  6. Select a Dataset ID.  In the case of this example, the Dataset ID is "person_id".

  7. Select the name of the Sample Date column.  In this example it's "Time Periods".

  8. Select the Estimator Target.  In this case it's "is_future_terminated".

  9. If you want to override any of the defaults, you can do so in the various sections under the Estimator Target

  10. Scroll to the bottom and select Create.

Run the Augmentation and review the results

When the augmentation is run, your data sources are fed into One AI where the features are cleaned and transformed and then many variations of many algorithms run against them.  One AI will then choose the best algorithm variation for the task at hand and the most predictive features and will create reports displaying the results. 

  1. To run, select the Run button for your augmentation from the Augmentations screen.

  2. You can view the progress by selecting the Runs button.  There are various statuses that the augmentation will progress through.  Augmentations can take anywhere from a few minutes to several days to run, depending on the data and configuration.  An hour is an average amount of time to wait.  When it's complete the status should say "Pending".

  3. Once the augmentation has reached Pending status, you can view the results.  To do so, select the specific run from the Runs screen.

  4. The augmentation results will open to the "EDA Report" (Exploratory Data Analysis) tab.  The data available on this report as well as the "Modeling Report" mentioned in the next step will provide you with the information you need in order to tell whether the model performed well.  If you're interested in learning more about the EDA Report, here is a help document that covers it: EDA Report Introduction

  5. Another report for determining whether the model performed well is the "Modeling Report".  Select the Results Summary tab to get to the Modeling Report.  We are not going to cover how to utilize this report here but the following help article is available: One AI Modeling Report Introduction

Re-configure and re-run the augmentation (optional)

If you're not satisfied with how your model performed, you should make changes to either the data that feeds your model or the augmentation configuration and re-run it.

To make changes to the data:

  1. Navigate back to Data > Destinations.

  2. Select the name of your destination to expand it.

  3. Select the gear icon next to any of your files and select Modify query in Explore.

  4. Modify the query as desired and run it.

  5. Select the Pin icon above the results and Update data destination query.

  6. Repeat steps 1 - 5 for each of the three data sets.  Again, all three should match in content and structure.

  7. Run the destination.  To do this, navigate back to the Data Destinations screen and select the Play button next to the destination you created.

  8. You must also refresh the augmentation in order for the new data to be utilized.  Navigate to Data > Augmentations.

  9. Select the Edit button for your augmentation.

  10. Select Refresh data from selected data destination.

  11. Scroll to the bottom and Save the augmentation.

  12. Re-run the augmentation and review the results.

To make changes to the augmentation configuration:

  1. Navigate to Data > Augmentations.

  2. Select the Edit button for your augmentation.

  3. In the settings sections below Estimator target, override any of the settings you wish to change.

  4. Scroll to the bottom and Save the augmentation.

  5. Re-run the augmentation and review the results.

Make the results available to report on in One Model

The One Model team can help you model and create metrics and dimensions from the data that One AI generates.  For example, you could add a tile to a storyboard in One Model that lists the departments that are likely to experience the most turnover this year.

  1. Confirm with the team at One Model that there is data pipeline logic in place to process the results from One AI augmentations.

  2. Navigate to Data > Augmentations if you're not already there.

  3. Select a specific run from the Runs screen for your augmentation that is in "Pending" status.

  4. Select the Results Summary tab and the Deploy button at the bottom.

  5. Navigate to Data > Loads.

  6. You should see a row that says "One AI data load!". Refresh the page until the status of that row is "Completed Without Processing".

  7. Wait until the data on your site is re-processed.  The frequency of re-processing varies but is generally daily.  If you would like the data processed sooner, please notify the team at One Model and they may be able to kick off re-processing.

  8. The metrics and dimensions created from your One AI data generally reside in the sections titled "Machine Learning".

In summary 

There are a lot of steps necessary to create a model in One AI and we haven't even gotten into fine tuning.  That said, once you have one created, each subsequent model you create is a lot easier.  Rest assured that the One Model team is always available to directly assist you in all aspects of One AI.

Did this answer your question?