[bitL]

No problem. I usually learn a new language by picking a very challenging idea and then implementing it; that way I can see how the language performs "under stress", though I might have skewed view for normal use cases. Still, companies seem to be more interested in the weird cases I am forced to explore (much to detriment of my happiness), and JavaScript was making me quite grumpy at times.

Right, with TensorFlow.js (or even better Keras.js) you can write a simple CNN in a few hours that given a good dataset of facial keypoints can allow you to build a fairly accurate (80-90%) emotion detection framework, and it's pretty fast. I am past this stage though; right now I have e.g. a working state-of-art model for detecting offensive visual content on the web based on very deep DenseNet, which might be difficult to fit into a browser implementation and the more Titan Vs you throw at it, the better; Python only then.

I agree, with functional approach, especially with the recent additions to ES, it seems like JS is becoming a quite OK language, though the issue of having somebody using ugly old hacks is not going away, unfortunately.

Usually in teams I worked before for anything more complex we used transpilers from one language to JavaScript and relied heavily on our IDEs to allow us debugging. Kotlin seems to be pretty popular in that space right now; if I haven't had some personal gripe with its authors I'd have probably used it as well ;-)

Good luck with your company, keep your eyes open for things to come! :)

[kapv89]

Thanks!

And by the way, using python for training is a must right now(though JS is going to become an alternate for this in a few months, as tensorflow.js is being ported to node with bindings to actual tensorflow's C++ layer). Using tensorflow.js, I'd have been able to use saved models in the browser.