I’ve signed up for Google Cloud Platform, and am working through a Udemy course on how to actually use it to run ML apps in Python. I find it all a bit intimidating, not sure why. My main motivation is to access more powerful resources, now that my needs are more demanding. I ran my trading app 50 times over my large dataset this morning and it took over 3 hours. That’s not such a big deal, except that I’ll be wanting to iterate over multiple variations (as in hyperparameter tuning) and that becomes unworkable on my local machine. I could upgrade, but there are still limits.
GCP gives a newcomer several hundred dollars credit on signup with 30 days to spend it. I had better not wait too long to actually give it a good try out. My trials (on my own machine) indicate that using a GPU is no faster than a CPU, but apparently this depends on how big the dataset is, and particularly how many features are being calculated at once because that’s where the parallelism of the GPU comes into it’s own. I’ve currently only got 24 features, and the 4 cores on my CPU are running at about 70% when I’m running my trading app. I doubt one of those fancy new nVidia GPUs would improve on that too much. So I’m not sure that GCP will actually speed up my code but there probably is some way I could achieve that.