Azure - while Auto-ML imbedded in Power BI can deliver standard classification and regression, it does require a Premium subscription. In a multi-tier model, Pro tier won’t get Auto-ML yet. If Microsoft’s stated goal is “democratizing ML”, Azure shows it can be done by 80/20 rule - 80% of the time, suffice as first quicklook. Just as features normalization from Lexus to Toyota over time, tiered features to Pro will be a welcome change in Q4.

AWS - while Alexa can play nice with Python, the process is anything but easy. Like the early Ford Model T with hand crank starter, it is a true test of endurance and skill. Understand Amazon has acquired the outfit that develops the Flask-Ask framework for quietly marrying Alexa and Python. Possibly (hope) we get a well integrated and fine-tuned offspring development kit, so we simply turn the key, start the Alexa-Python engine running (no hand crank). In the same vein, as hassle free as scripting Alexa’s “utterances” and “intents”.

GCP - what may lead to a step change on seismic interpretation using unsupervised deep neural network. It takes the laborious work of interpreters / geoscientists labeling the geologic features out of the time-resource equation. Understand the architecture is a mutant variant of ResNet, with order of magnitude different breadth and depth (thousands of neurons / layers). So TPU may be just right for time-critical applications of Ultra Deep Seismic Network (pre lease sale prospecting to quick subsurface reservoir evaluation).

ResNet easy read: https://towardsdatascienc...resnets-c0a830a288a4

Food for thought - harnessing real-time power, what if 737 Max 8 uses Tensorflow with TPU?