Thank you to those who attended Owais Hashmi's data science meetup in Miami during
IBM Analytics University and joined the
IBM Data Science Community. Check out some quick notes from Owais talk below and please join the discussion by leaving your comments and feedback.
Data Science & ML Software Adoption: According to KD Nuggets, Data Science & ML software adoption currently leans heavily on Python and R.
IBM Watson Studio quick highlights: Data science teams have typically worked in silos; in Watson Studio, we've incorporated a heavy element of collaboration, placing a high priority on the ability to connect to data sources in a way that's secure and governed.
- Offering 34 connectors with the ability to support Google, Amazon and other open source databases, we major on metadata, allowing data scientists to understand what they have and where.
- In Watson Studio, the dashboard will first ask what you're trying to do: refine or clean the data? Run a machine learning algorithm? From there we've integrated a recommender component with Watson, which allows you to see data sources that have been used in other projects by members of your organization. You have the option to accept or reject Watson's recommendations.
- The search mechanism will pull up any .pdf, connections going to database sources, tables, etc. related to any query, i.e., "insurance." Users will see the schema of the dataset, with certain columns anonymized so all of the data is surfaced in a way that can still be usable. It'll also search data related to "insurance" as it learns from other user searches.
Model building with Watson Machine Learning: Users can easily access data and build models while up to 10-20 different services run underneath.
Studio supports Model Builder, New Editor and Notebooks. You have the option to build a model manually or have the system build one for you automatically.
Example: some data scientists have immense domain knowledge, knowing their data inside out, but don't want to write code. We built an intuitive model such that when you point to the data, Watson uses a guided approach to suggest a technique that might build a good model. A second approach is to independently build your own visual models. The third is the fully guided approach, in which you ask the system to build a model for you. It does this by giving you the results of all of the algorithms, letting you make the final decision on which one to use.
You can set up your model on a schedule so that it recalibrates and retrains at intervals that you specify. You also get a view of estimators and accuracies so that you can see what you want to deploy against.
Deep learning - what's the best framework? We went with the agnostic approach; what if we could build a deep learning model without specifying what language? We use the neural network modeler, going one level higher in the sense that what you would have built in code is actually built visually. You either use the cloud to train against your network or you take the raw code and go behind the firewall, giving you the flexibility to get out of the system and download to TensorFlow, for example.
Issue of bias. This is a huge barrier to delivering business value using AI in applications. Recently, we've seen AI models that do not recognize individuals with certain skin color, ethnicity, age, etc. With so many models that are being built, why aren't they being adopted in business?
They are not trusted. IBM's Trust & Transparency capabilities for AI on IBM Cloud aims to tackle this.
What other questions do you have about AI in applications, Watson Studio, or other Data Science tools and programs? Leave your comments below.
As a Data Scientist in the leading IBM Watson Data and AI practice, Owais Hashmi assists clients on their journey to become data driven organizations by solving complex business problems leveraging the best in class AI technologies. Having engaged with customers across various industries globally he brings a fresh perspective to the field of AI and enjoys sharing his insights at tech conferences and events.
#GlobalAIandDataScience#GlobalDataScience