lifeslogit

I'm pretty sure most of us DS know about significant digits and are usually calculating the maximum to enable future flexibility. For a single output, I can understand how you'd be upset we don't round. But for the 20+ column tables we usually build, I've found most will calculate the maximum scale allowed by the database and call it a day. The best of us certainly find the right formatting for the presentation.

In fraud analysis, we look for these red flags! Fraudsters come to the office earlier and stay later than others. They take on as much as possible to expand their access and take no vacations to make it so that others never review their work.

Fraudsters looks a lot like a great employee! To avoid this, implement healthy segregation of duties. Trust is great but also the fraudsters greatest weapon.

In fraud analysis, we look for too much access (what we call improper segregation of duties). This is pretty common at smaller organizations, so it can help to look for additional red flags. Fraudsters commonly show up earlier and stay later. They rarely take vacations. The goal is to control as much as possible and avoid having others review your work!

For many years my partner and I operated in the same way. I paid our rent, utilities and food while she spent on her hobbies and saved. I was not unhappy during this time because I didn't want much. A few years ago though, I became a bit more financially aware after having my first soul crushing job and realizing I couldn't rely on work to produce income in the same "easy" way I had when I was younger. The emotional cost had become too high. Managing this part of our relationship continues to be a multi-year process requiring ongoing discussions of what we have, what we want and what we'd be willing to do to get it. It feels like a muscle that atrophies, but I have made my peace with that because it works for us. I remind her what she wants and how she can get it by helping me now or spending less now. I even ask her to provide the same feedback for me. Her perspective on my spending is as important as my perspective of hers. I imagine we will regress in the future. Those moments will probably suck and cause a lot of stress. For now my only advice is to make a habit of these discussions in your relationship and protect the habit as long as you can.

That is very interesting to hear! I imagined that this was the perfect storm for HFT. Lots of retail investors using default exchange routing settings via brokers like Robinhood and ETrade that sell their order flow. Meanwhile these retail investors are buying small amounts of very volatile equities! Lots of opportunity to rack up lots of pennies.

The tragedy of the commons is a great launching point for understanding this problem. Managing fish populations is one of the best examples.

For further reading, I suggest folks check out Elinor Ostrom, an economist who spent time developing regulatory solutions for managing fish populations. She was even able to implement a few!

For a researcher, this difference can lead to deep unhappiness. I moved from a research-heavy institution to a founder-heavy culture thinking the freedom and increased salary would lead to improved happiness, however this was very far from the case. After about 1 year, my CEO began to understand the difference and support me, however, the time and stress prior to that point was very difficult. It required Investor-level individuals with research careers to validate my perspective. Sam's post validates my struggle and I am happy to see it publicized by someone with clout. I hope more founders will begin to give researchers a bit more room and support.

Seconding this. fast.ai has greatly emphasized practical approaches for those interested in building deep learning applications. While I haven't completed their Deep Learning from the Foundations course, I don't know of any other course that goes as in-depth on the fascinating topic of building your own Deep Learning API.

Do whatever it takes to get you in the working mood. All tips aim to solve this abstract problem. I used to move all the furniture in our NYC apartment and set up all my monitors, then take them all down and move it all back at the end of the day. Snacking was a problem so I had to meal prep. The funny thing is I don't even meal prep while at the office, but at home I needed to add that structure. Simply put, identify the many dimensions along which you suck at and implement robust rules, blockers, systems or penalties to save you.

There are layers of each problem space that benefit from optimization and good design more than others, however, most solutions still need all the other layers to be useful. I've had extreme success being a data generalist at a mid-sized analytics company because I was able to be the one to fill in all the gaps of a project, e.g if there are 10 steps that need to be completed, the generalist can do 7 of them and the specialists can do a great job on those other 3. This is the magic of human collaboration after all! On the flip side, I've found that working alone I miss having the specialists. While I know I can technically achieve all the steps, what I really want is the context the specialists can add. Those bits of code or math that turn a vanilla engineered solution into a nuanced piece of software. My conclusion is that collaboration between generalists and specialists is important but asymmetric. Generalists need context and information from specialists so they can become semi-specialists, while specialists simply need generalist infrastructure to power their ideas.

Some other ways Brokers make money: Market Making operations that leverage the same infrastructure and market position, Interest on Margin, Stock Loan/Borrow activity with other Brokers, and Market Data Access.

Still with these items, I agree that a move to 0 commission trading will cause a shift.

We keep it very simple as well.

Postgres read-replica for cheap realtime stuff, Redshift for the larger sets, Airflow to manage transfers, and Metabase to visualize and share. We also collect data from a bunch of APIs, but those each run via their own job and store in Postgres.

We also try to define wide short tables in Redshift that enable users to use Metabase to ask questions.

I was very happy with Metabase. Being that we can't afford Looker right now (but we would all love to) it is pretty solid.