Test, quizzes, and other measures of knowledge.
My weekly accountability report for my self-study approach to learning data science.I'm still feeling positive about the Full Stack Deep Learning boot camp. I check the slack channel with alumni postings almost every day, which I'm not usually a Slack person (or more accurately haven't felt a need to be a Slack person previously) so that probably says something about my level of connection with the group.
And I've been studying for the FSDL Alumni test which I hope to take by the end of the week. This feels like the same sort of studying I would do to prepare for an interview so it's an interesting use of my time, shifting from learning things, or attempting to produce content for portfolio reasons, to a "can I convince someone I know stuff." There are lots of different modes to work through in trying to become a data scientist, and this one is a bit new to me.
My approach is to cover and review my notes from the Coursera Deep Learning Specialization as well as the lecture slides from the boot camp. One thing I've noticed is that my memory is not what it was the first go-around in graduate school. So that's fun. I'm glad I took mostly good notes; I did go back and redo some of the convolutional ones where clearly I was tired and my basic math is not checking out. I'm reviewing things like activation functions, initialization of weights, CNN and LSTM architecture, and structuring/evaluating projects (which I realize I really, really enjoy).
This time, I'm transcribing the most salient points into questions into flashcards which I can review in 5 minutes bits of time when I don't have the space to do deeper coding or learning projects. It's interesting that while I draw knowledge from many sources (blog posts, MOOC videos, books (online or otherwise)) I do need to have them in one central spot in order to review/study them. There's a lot I don't like about the Anki cards, but at least they are all together.
I also tried out Workera, a deeplearning.ai company, that administers online tests in the ai/data science space. For self-directed learners I highly recommend it. For jobseekers it's free. Take a handful of tests, get graded, then you are given a course of study to improve and an analysis on what role might be good for you along with a list of job postings and direct referrals that you qualify for.
The test areas are machine learning, deep learning (opt), data science, mathematics, object-oriented programming, algorithmic coding, software engineering (opt), and communication ability (opt). The communication ability is a one-way video call, the algorithm coding is a python
coding environment, but the rest are multiple-choice questions. Once you complete them all they will suggest a primary and secondary role in one of the following: Data Analyst, Data Scientist, Machine Learning Engineer, Machine Learning Researcher, Software Engineer - Machine Learning Engineer, and Software Engineer.
It recommended data scientist as the closest role based on my scores, followed by a data analyst. My top scores were in machine learning, data science, and Software Engineering, while my deep learning, mathematics, and algorithmic coding scores were all low. But it also included study guides (yeah! tell me what you think I should study and maybe, I will and maybe I won't but at least I have some input!).
I took the tests cold and didn't do so hot, which I was expecting as I've done very little interview prep so far. The algorithm coding included basic things like tree traversal and honestly, it's been over twenty years, so I'm not surprised at my results. But I am excited by some actual concrete feedback and where I'm at and what I've been doing for the last year. I can take the test 2 more times, then try again in 90 days. This works well with what I'm trying to focus on between now and the end of the year, with the plan to applying for work in Q1 of next.
There are a few job postings listed (a dozen for San Francisco) and if you meet or exceed the company's scores you are eligible for a direct referral. As an entry-level person, anything that allows me a leg up over cold applying on LinkedIn or a company website seems like a great idea. Here's an example of a radar chart for a position listed as a Data Scientist (that Workera classified as machine learning engineer role), showing what the company's performance requirements are in green vs umm, some blue dots that most definitely aren't my scores.
This is nice even if it gives you an idea of what the recruiter
thinks is the relative importance of a set of skills. That being said there still could be a mismatch going on.
My general impressions are really positive even if it's still a bit rough around the edges. I'm not convinced it's a perfect example of what people in those specific roles should know, so as a self-taught individual I still take complete and total responsibility for what I'm learning. And as someone who has opted out of traditional schooling on numerous occasions and side-eyes any standardized test, I'm probably not going to be thrilled if this somehow becomes an industry standard.
Also, I'm new to things, but there were a few things that didn't jive with my experience of the industry. For example, I was a bit surprised at the data science questions that were asked (it seemed focused on more basic probability focused than I was expecting). There were no questions about SQL (that I recall?). I completely bombed the mathematics portion which was mostly linear algebra, a dash of calculus and some functional analysis where I choked on notation. I have been reviewing these within the context of deep learning and machine learning but maybe I need to be able to answer them as stand-alone interview questions. Also, it seemed engineering heavy with a lot less statistics than I was expecting. I'm guessing that this reflects the background of the deeplearning.ai folks who seem to be coming at it from an engineering and research angle. It certainly doesn't match my experience of professional data science twitter for example, or maybe I just follow a lot of statisticians.
But despite those, I am thrilled to have this asset in my self-directed learning toolbox. If you've used this site, I'd love to hear your experience.
Top comments (2)
Cool, I'm gonna give workera a look. It might be a form of procrastination, but it also might be a good use of time. I will be interesting to get a snapshop of where I stand.
Yes! And let me know what you think. One of the things I struggle with is where do I want to be on the analyst to engineer spectrum and this is helping to give me some context to figure that out.