I heard about Azure Machine Learning when a client wanted a demo of the product to solve their machine learning needs. I started my research into the different cloud machine learning solutions that different companies provided, companies ranging from Amazon, databricks, Microsoft, Google, and IBM. I found that every company wanted to cater a specific ease to their users and Microsoft Azure Machine Learning particularly catered to me visually as a data scientist. It seemed almost fun to be able to drag and drop big chunks of the data science workflow and connect modules together which felt like I was able to see the big picture of my train of thought at every instant.
I started my learning at this Microsoft Virtual Academy link:
Started my free trial and worked out the material:
explored the tutorial within the sample experiments section:
Glanced at the Gallery of solutions ready to be viewed:
Finally started to build my models:
I started building my experiments with the data provided for me by my client and found that I was able to perfect my model faster than usual because the following qualities that I think make this technology stand out from my point of view.
Visual and mental clarity – at each step I’m able to see my train of thought leading up to what I’ve done instead of seeing a vertical list of code that I need to scroll through to remember where I left off, not to mention making changes to your train of thought.
Insanity checks, targeting error – being able to see your dataset at each module visualized for you and model evaluated visually for you is immense. You can see the statistics and graphs that help you keep yourself on top of how the data is transforming and assessed which is great for debugging later and makes you accountable not just to an individual performance metric but to the methods used to get to a good prediction.
Adaptable – has diverse types of databases it can import from and if you want to use SQL/R/Python for little pieces of code that you want to add to the model you can. You can work on the same model with multiple people if you share your model. This increase usability among the data science community and encourages collaboration.
Hammers the agile approach – you can deal with unpredictability through incremental, iterative work and empirical feedback.
Helps you learn datascience faster – I felt that this would be a great educational tool for people that have never tackled the messy world of analyzing big data. It makes all the freakish sounding algorithms into drag and drop modules, can structure your work flow smoothly to match the crisp-dm methodology (the scientific method of datascience), and ideal for visual learning of the evaluation of a prediction or dataset. Overall, brings back interpretability of what it’s all about and hammers fundamentals quickly.