I somewhat did what OP is looking to do, albeit at ~33. I also had a PhD, but to most tech people the fact that it was in psychology is all they see, they aren't really aware there are quant/stats heavy sub-disciplines. I moved from a project manager in HR to a research scientist in tech within the same fortune 500 company and then transitioned that to a principal data scientist at a consulting firm. I was even offered the job of a director of data science in the tech division, over a team of 25 at another fortune 500 and turned it down. I have recruiters from Facebook and Amazon, etc. Reaching out to me very regularly, but for some reason they still want you in Cali or NYC 🤔
Just offering another perspective. Data Science is such a young field very few have traditional backgrounds and it's something you can bring to any role you have.
For example, my wife works in finance and she told me somebody she works with manually looks up over 100 mutual fund tickers on Morningstar every month to get the morningstar rating. Just for fun in like a half hour I built a script that takes in the ticker, generates the URL, goes to the URL, scrapes the html, and saves the star rating into a spreadsheet for them. So they no longer have to spend hours looking up the tickers every month. Another example, when I was a research scientist our company paid $15k a year to outsource the manual coding of open-ended surveys we conducted. In an afternoon I took a few years of previous labeling and build an ML algorithm that predicted the class with 99% accuracy. Eliminating the need for us to pay the $15k/yr to have them manually coded.
So much automation and data analysis that could be done in sales to get your feet wet and see if it's something that may be worth pursuing further. Could you generate a script that sends you a weekly email letting you know the last time you contacted each of your clients, etc.? These are just ideas to where you could integrate programming into your current job, so when you try to get a new one you can use those as "work examples" of programming experience.
Domain expertise is extremely important in data science too.
This is fantastic advice and also somewhat similar to what I did...I went from a completely unrelated field to data science and started out by automating manual reports in my existing job, which freed up a ton of time for me to read tech manuals while I ran my programs. There are so many jobs which rely on manual spreadsheets that have to be tediously updated on a regular basis, for example. Even just learning how to record a macro in Excel and then stepping into that macro and looking at how the vba code works and making little tweaks to it is a starting place. You'd be surprised how much you can automate once you start looking for inefficiencies.
Funny story with the mutual fund automation was that I originally couldn't find the star rating in the source code because it was an image, so instead I built a convolutional neural network based off of Resnet34, where the script went to the website took a screenshot of the section of the page that contained the stars, and put that image through the model to get out the prediction. I'd tested it on over 40 and it was 100% accurate, but when I found the reference to the number of stars in the html and that I could easily scrape that I figured that was a bit more foolproof.
I agree so much automation to be done in many jobs. Plus I'm sure in their CRM there is tons of data that nobody is even looking at. I know at our company outside of basic stuff none of our sales team uses the data. You could build clustering algorithms to identify like customers, build customer retention models, flight risk models, etc. A wealth of data in most CRMs that almost nobody looks at.
And in my experience the pay is good. I had already made pretty good money before, but in ~5 years since making the transition to DS I've doubled my income and I work remotely.
What is a good way to learn the skills to get started in DS at an older age?
Basically what I did was tried to integrate what I learned into my job. I feel like that really helped hammer home the concepts. But as for courses I found all of these very helpful. Also, finding data you are interested in can be helpful too. I like baseball, so when I have free time I use the pybaseball api to download the statcast data and try what I've been learning, etc. with that data.
Very solid introduction into Python:
https://www.udemy.com/course/complete-python-bootcampTougher, but intro to python and computer science principles:
https://www.edx.org/course/introduction-to-computer-science-and-programming-7intro to python's numerical library (numpy), their dataframe library (pandas), their visualization libraries (matplotlib, seaborn, plotly, etc.), and their machine learning library (scikit-learn):
https://www.udemy.com/course/python-for-data-science-and-machine-learning-bootcamp/fairly easy intro to deep learning via fastai and Jeremy Howard:
https://course.fast.ai/Jeremy's new book:
https://github.com/fastai/fastbookI'd say the first 3 will get you to feeling competent after that I'd focus on finding something that inherently interests you and trying to use what you learned to manipulate the data and look at the results.
Another very useful tool IMO is Kaggle kernels, just to see how other people approach the same data and datasets. For example in the python for data science and machine learning bootcamp Jose uses several different datasets for his weekly lessons, like the Titanic dataset. You can go to kaggle and see tons of analyses, models, etc. built on the Titanic dataset just to see how other people are approaching the data.
https://www.kaggle.com/c/titanic/notebooksFor example this person wanted to leverage the k-means clustering algorithm to investigate the data:
https://www.kaggle.com/ryati131457/titanic-unsupervised-kmeansIt's definitely not easy, but IMO learning about this stuff was interesting and like I said I'd quickly try to integrate it into my work by thinking about problems at work that the things I've learned could solve or help solve.
I still am reading Arxiv articles and taking courses today years later because this stuff all moves so fast.
This is the fun stuff we're investigating at work right now :)
https://arxiv.org/abs/2005.07683Hopefully this is helpful, let me know if you have any other questions.
I also forgot to mention those udemy links can be dynamic, don't pay more than $15 for any course on udemy if you do decide to pay. If they don't appear to be less than $15 wait a few days and they will be :) If you get impatient you can shoot me a message I think I have codes from when I've paid for Jose's courses that knock it down to like $9.99 or something around that.