Show more

I've been heavily refactoring code of my machine learning project. Now I can just watch how GPUs are fully loaded, while the CPUs are not that much. Can't wait for first validation results. It's a slightly modified model which takes images of different sizes. I don't know yet wether it gives better results, but it definitely heats up the room.

I was sluggish this morning, but something is wrong if one is active at 6 in the morning. I was late to the pool for a swim practice, and fast people took over my usual lane, as a result, I felt like a gazelle chased by lions. Anyway, I survived it.

The one I had was too short, now I got a cable which is long enough to reach my home.

The morning started by putting a new Ethernet cable in my office.

The server crashed shortly after I left, so no great results.

vmtouch is a nice tool to cache file content into memory.

Then I relaised, that resising images on the fly might be the main reason of the slow down. This idea came to me on Friday afternoon. The end of the day was intense, but resizing images once and staving the sizes I need improved the speed. Still, the GPU is not 100% busy. Now I'm looking for Monday, to see the results!

At that point TemsorFlow was complaining that the CPU supports instructions that the binary was not built with. Before building TF from source, I decided to update cuda and other Nvidia software. After a dance following outdated tutorials, I got it done. Speed improved, but still, while CPUs were at 100%, GPUs were barely loaded.

When I moved my code to a much more powerful machine with lots of memory, cpus and gpus, I got a GPU enabled TemsorFlow binary. That speed up the things, but still CPU was a bottleneck.

Initially I worked on my laptop with an extremely small values, for example, images were resized to 128 by 128, which feels too small. However, I managed to get reasonable performance numbers.

I picked up Keras, so the code for the network itself is simple. Complexity is around: dataset preparation, evaluation and model selection.

I've started a machine learning project using TemsorFlow. The first thing I've learned is to start small, implement all the functionality and only then move to a real task.

As an experiment to use Spotify less often, I'll give somafm.com a try as the default radio in my car.

A big plus is that I won't listen to the same music over and over again. The only worry I have is that I'll pay for the data about what I'm paying right now for Spotify.

It's also a good time to donate to somafm!

I start thinking of large IT companies as drug dealers. It really helps using social media less often.

Usually, a paper is mostly done once I get an appropriate plot to show the results. This time I feel I've came up with a good title, or at least a section name, that distills the argument.

I must be the only person ordering macchiato on my way to work. Every time I make the order, the reaction is yeap, macchiato, that's Dimitri.

Show more
3dots.lv Mastodon Instance

A beta setup of a Mastodon instance primary for family and friends.