Notes

Install Linux VERSION=1.8.0 WORKING_DIR=/tmp/syncthing-$VERSION rm -rf $WORKING_DIR mkdir $WORKING_DIR cd $WORKING_DIR wget https://github.com/syncthing/syncthing/releases/download/v$VERSION/syncthing-linux-amd64-v$VERSION.tar.gz tar xzvf syncthing-linux-amd64-v$VERSION.tar.gz cd syncthing-linux-amd64-v$VERSION # If the following command throws an error like: "cp: cannot create regular file `/usr/local/bin/syncthing': Text file busy", # shutdown the current Syncthing session from the UI and retry. sudo cp syncthing /usr/local/bin Note: Change version number to the latest one. Mac brew cask install syncthing If you run into any issues (for e.g., Syncthing doesn’t start), try starting the daemon through the command line to see what the problem is: ...

Statistics While the following from statistics is too fundamental to be interesting, I struggled (and failed) to fully grasp it a few years ago. Hopefully, it will stick this time! Anyway, before I forget: Standard deviation is the square root of variance. Variance is the average of squared sum of differences between mean and observed values. Standard deviation is a measure of how spread out observed values are. Normal or Gaussian distribution is simply a bell curve, something that is fairly common in the real world. For such a distribution, 68% of the values lie within 1 standard deviation of the mean, 95% within 2 and 99.7% within 3. Standard normal disribution: is a (special) normal distribution which has a mean of 0 and standard deviation of 1. The y-axis of such a graph is just numbers but the x-axis is called the Z-score. You can convert a normal distribution to a standard normal distribution by calculating the Z-score as: ((observerd-value - mean) / standard-deviation). The area under the curve is 1. If you search online, you’ll find a table that, for a given Z-score, gives the area of the curve on the left of that score. That will answer things like, what’s the probability that a given value X is less than something. Resources: ...

Fastai: Initially, when you start training, accuracy continues to go up. However, the model will start memorizing after sometime and validation set accuracy will decrease. That is when the model starts to overfit. The model looks at all images exactly once in every epoch. Architecture (for e.g., ResNet) is a template for a mathematical function. When trained, it turns into a model which contains parameters, possibly in millions, appropriate to tackle the problem at hand. Train = fit.

I began part 1 of the Fastai course sometime in Q4 2019. After an inordinate amount of time for something that I could and should have finished in about 3-4 months, I git pushed my last notebook early last week. Nevertheless, I was excited but then Fastai launched a newer, and possibly a more improved, version of their course a couple of days later! The new course is apparently better because it incorporates some aspects of traditional Machine Learning (as opposed to how the previous one was only about Deep Learning) and is taught with a newer version of the Fastai library that they launched alongside the course. ...

Ebey’s Landing National Historical Reserve: looks beautiful. Bluff trail (3.5 miles). Whidbey Island Bagel Factory: it was pretty good. There is one beach on each side of the ferry terminal - can visit those next time. Not do: Fort Casey’s: I’ve already been there twice, so try to find something else next time!

Motivation It has been quite awhile since I’ve been trying to figure out how I want to operate at work. The company expects me to be more of a tech lead because that will allow my manager to scale out and get work out of a larger workforce. And while that does have some merits, it isn’t the best way for me. It isn’t fun and I don’t get to learn much. It isn’t what other companies expect from SDEs, so it decreases my employability. (For example, Facebook and possibly both Google and Microsoft expect their senior engineers to contribute a lot of code). Even within Amazon, they’ll look at my deliverables at the end of the year. If I only work through others, I won’t have a lot of tangible stuff to show forth. I’ve also heard from my manager that whatever I do inside the team isn’t going to take me to the next level, so why do that as much? The other side of the story is that not behaving as much of a tech lead could lead to some reputational loss, but I think that’s an acceptable tradeoff. ...

The following will query CloudWatch Logs Insights for last 24 hours. If you want to search for a different duration, adjust the --start-time and --end-time parameters accordingly. queryId=$(aws logs start-query \ --log-group-name '<log-group-name>' \ --start-time `date -j -v-1d +%s` \ --end-time `date +%s` \ --query-string ' fields @timestamp, @message | sort @timestamp desc | limit 20 ' | jq --raw-output '.queryId' ) aws logs get-query-results \ --query-id $queryId

There are two ways of doing this and they both work with Git submodules. git reset --hard <commit-ID> Obviously risky as it will overwrite all local changes in the working directory. Works better when you only depend on Git submodules but don’t make any code changes to them (such as Hugo themes). Or: git checkout <commit-ID>

Pytorch: x += 1 and x = x + 1 are not equivalent in Pytorch! The former is in-place and the latter is not, so be careful. More explanation is available here: x = x + 1 is not in-place, because it takes the objects pointed to by x, creates a new Variable, adds 1 to x putting the result in the new Variable, and overwrites the object referenced by x to point to the new var. There are no in-place modifications, you only change Python references (you can check that id(x) is different before and after that line). ...

We are expecting our first baby soon, sometime in mid November! When I learnt about this, it took awhile for the feeling to sink in. I am obviously super-excited and looking forward to when the kiddo arrives. I am also intimidated by the unknowns: I’ve heard some rumors but will only understand them when the time comes.