I have some research I'm looking to do on time-series in python in the coming weeks. In the longer term I do want to understand the math/stat underpinnings (I'm using Brilliant to catch up on maths), but, for now, could I make decent predictions/tests using python libraries? Perhaps by reading a layman's explanation of various outputs first (math is the limiting factor in this timeline)?
Probably, but it's almost certainly not a good idea.<p>This is a pretty good (and free) textbook: <a href="https://otexts.com/fpp2/" rel="nofollow">https://otexts.com/fpp2/</a><p>To elaborate on my point above, you can definitely get some results quickly without understanding how it all works, but even a small bit of knowledge about whatever method you're using can help dramatically when you're trying to debug things.<p>I don't think the book above has that much maths, and it's definitely aimed at newbies to forecasting. The examples are in R though, which may not be particularly useful for you.
Tons of companies have and are hireing people for little more than beeing able o download install and run the pytorch examples, and who have possibly managed to retrain an existing network architecture for a similar task on a similar dataset. They are absolutely the script kiddie equivalent of a data scientist.<p>A simple trick question to see if you are one: Given a dataset with significant sample imbalance, how should the optimization be adjusted to account for this?
Maybe. What's your question more concretely?<p>If you want to just figure out how to call some functions and get an output from them - yes, of course you can. Do you care about getting useful results? Do you want to do this as a full time job or is it just a one-off time series problem?
Can you produce really insightful and correct analysis of time series without understanding what you are doing? No.<p>There are no shortcuts with statistics. It will all seem to work out fine until you involuntary shoot yourself in the foot and you most likely will because doing forecasting properly is tricky. If you are lucky you will notice something doesn’t look right but you might not.
So there's <a href="https://fast.ai" rel="nofollow">https://fast.ai</a> MOOCs+library which don't require a big theoretical background<p>I heard about a program that looked great to do data science from just data a while ago, but I can't find it :/ It was on Show HN I think<p>I found that tho, don't know if it's gonna be helpful:
<a href="https://towardsdatascience.com/top-8-no-code-machine-learning-platforms-you-should-use-in-2020-1d1801300dd0" rel="nofollow">https://towardsdatascience.com/top-8-no-code-machine-learnin...</a>
<a href="https://www.obviously.ai/" rel="nofollow">https://www.obviously.ai/</a>