As in machine learning, lets define Geek meter 1-10 (max out , 10 all command lines, scripts, bash etc.;1 - drag-n-drop with minimal coding). Here is a snapshot from the perspective of a practitioner with deployment as the goal.

If much time-resource is spent on collaboration, workflow design and enrollment of practitioners from different disciplines, Azure machine learning (ML) Studio is a natural fit with data processing hand in glove. Highly intuitive. Low on the Geek meter.

If the focus is on using TensorFlow and latest TPUs for computing supreme, then GCP is a solid start. High-level TensorFlow API (estimators) hides the complexity of setting up number of layers and neurons.While TensorBoard is great for computation graph visual, it is also high on the Geek meter.

If the goal is usability and reach to as many stakeholders (say from boiler room to the board room), AWS with Alexa voice front end, as browser to the internet, coupled with lambda make a compelling platform. Right in the middle of the Geek meter.

So it depends on what we want to accomplish and deemed good value (e.g. why we do ML and what is the problem we want to address, e.g., return or volume).

Nirvana? ML Studio workflow design, high-level TensorFlow Engine + Voice interface.

Witness rate of innovation: external open-source > internal org-specific, we are getting there fast.

Free free to chime in, share your cloud ML experience, and make AAPG a truly learning community.