Automate the deployment workflow with Watson Machine Learning
Last year we made data science a team sport with IBM Data Science Experience, our award-winning IDE for analytics. This summer we brought to market IBM Watson Machine Learning that allows companies to put models into production with easy model management and full workflow automation. And last week, we announced we’ve grown up those two products into Watson Data Platform, while adding new features. Following are some examples:
Continuous learning – your models should always improve
Models today are difficult and time-consuming to maintain and to keep always up to date. With Watson Machine Learning, it’s possible to automate the retraining of models and to monitor how the performance of those models evolves over time — that’s what we call continuous learning, which is a unique feature in our platform. Thresholds can be setup and if the performance drops, the user will get alerts and notifications so the data scientist can act.
Version control of ML models – keep track of all the changes
With Watson Machine Learning, we provide complete lineage and governance of those models, specifically for audit purposes. Models are dynamic assets that need to be updated periodically. That’s why it is key to have version control and the option to roll back to previous versions when needed — and to have this all be accessible through APIs and the UI. Once you retrain the model, every model is saved as a new version, so in the “evaluation” tab you have all the model versions, in case you want to go back to a previous one.
This is important because in May 2018, the EU’s General Data Protection Regulation (GDPR) will take effect and grant consumers a limited legal “right to explanation” from organizations that use algorithmic decision making.
xgboost Model Deployment – more accurate…and faster!
xgboost has become one of the most popular open source machine learning frameworks in the industry. More than half of the Kaggle competition winning solutions use xgboost, because it is powerful, accurate and fast. Watson Machine Learning supports xgboost as a first-class framework with support of the entire ML flow (train –> save in repo –> deploy –> score –> automatically retrain). Here you can find a Jupyter Notebook tutorial to train and deploy a xgboost model.
Python client for WML — use your favorite IDE
Review PyPi, the Python public repository for Python Libraries, to see the new Watson Machine Library Python Library. The Python library is a wrapper on top of the WML APIs to make it easier for Python users create models and put them into production.
The library comes pre-installed with Data Science Experience but Python users can install the library in any Python IDE like PyCharm or Open Source Jupyter . To install it simply use the popular !pip install watson-machine-learning-client –> easy!