August 3, 2017 | Written by: Anthony Stevens
Share this post:
What is Watson Machine Learning (WML)?
After 12 months in beta and with input from hundreds of beta users, we are excited to announce the general availability of the Watson Machine Learning (WML) service. WML is designed to address the needs of two primary personas:
- Data Scientists: create machine leaning pipelines that transform data and apply machine learning algorithms to train predictive models.They typically use notebooks such as Jupyter, that is built into tools like the IBM Data Science Experience (DSX), to train and evaluate these models.
- Developers: build intelligent applications that leverage the predictions generated by machine learning models.
WML allows users to deploy and monitor machine learning models that were trained using SPSS, Scikit Learn and Spark ML.
- Training: the process of applying an estimation algorithm to ‘learn’ from a data set which generates a model that can make predictions related to that data set.
- Scoring: the operation of predicting an outcome using a trained model.
Automated Model Building
WML’s model building wizard is exposed through IBM Data Science Experience (DSX) to provide a guided workflow that walks users through creating machine learning models. Start by selecting to either build manually and choosing the algorithms best suited for your needs, or select the Automatic option and get help selecting the algorithm that would work best for your data. The Automatic option guides you through preparing data for training, and offers recommendations on algorithms to apply based on characteristics of your data.
Training models are not the primary value proposition of the WML service. Data scientists already have a comprehensive suite of open source tools to perform these training tasks. Instead, the challenge now is to operationalize those models. How can machine learning models be deployed into production to create business value? And once those models are in production, how do they adapt and evolve over time? These are the challenges that WML is intended to address. Once you’ve trained your model, WML helps you easily deploy to a REST endpoint that will automatically scale as needed. The starting point is WML’s tooling which displays all the models you’ve uploaded via the APIs or shared with WML via DSX. These models can be created in SPSS, Jupyter Notebooks, WML’s Model Builder UI, or imported from outside using your own model training tools as long as they export to one of our supported model formats.
List of Trained Models
Integration with Data Science Experience
Once a model is trained in Data Science Experience
, data scientists can easily collaborate with other team members (e.g. app developers) by sharing their models. Once shared, these models appear in WML and can be quickly deployed to a REST endpoint. And speaking of REST, WML provides a powerful set of REST APIs which are fully documented here: http://watson-ml-api.mybluemix.net/
Multiple tutorials are provided to guide you through the end-to-end process of training, saving, and deploying model within WML.
Additional screens guide users through subsequent model build steps.
- Upload your dataset followed by guided data cleansing
- Specify the target variable to predict
- Selecting the type of model to build (e.g. which machine learning algorithm to apply to the data).
Several tutorials are provided to guide you through using WML’s new Model Builder to create different model types. Once a model is created, it can then be deployed as in the next section.
Like the other Watson services, IBM Watson Machine Learning is available on Bluemix
, IBM’s open cloud development platform. If you haven’t already tried it — or you have and want to see how WML improves upon it — head over to IBM Data Science Experience
to check out this exciting new capability!
See you inside of WML and please provide feedback!!
IBM Data Science Experience team