Profile cover photo
Profile photo
Victor Costan
Coding Maniac, PhD student in Computer Science at MIT
Coding Maniac, PhD student in Computer Science at MIT


Post has attachment
The emotion wheel below is intended as a tool for the emotionally challenged to express their feelings. I think it's equally useful for ESL (English as a second language) folks. I don't remember most of these words showing up in TOEFL or SAT study guides.
Emotion Wheel
Emotion Wheel

Post has attachment
Got a Chrome refactoring award.

Post has attachment
Succeeded in debugging Chrome on Fedora with VS Code and the vscode-lldb plugin.

sudo dnf install -y lldb python-lldb

Launch configuration:

"type": "lldb",
"request": "launch",
"name": "Debug",
"program": "${workspaceFolder}/out/Default/net_unittests",
"args": ["--gtest_filter=CookieMonster/CookieStoreChangeTest/0.Named_Insert"],
"cwd": "${workspaceFolder}"

Post has attachment
C++ reference for Protocol Buffers. MessageLite has most of the useful message-level methods.

Post has attachment

Post has attachment
Colaboratory from Google Research is a hosted Jupyter environment where the notebooks are saved in Drive and can be shared.
Colaboratory – Google
Colaboratory – Google

Post has attachment
Commands I used to build Tensorflow CPU for Fedora. My custom build trains a few times faster than the version I pulled with "pip3 install" on an i7-6700K CPU (4 cores, 4GHz).

sudo dnf copr enable vbatts/bazel
sudo dnf install bazel
sudo dnf install python3-numpy python3-devel python3-pip python3-wheel
git checkout v1.4.1
bazel build --config=opt //tensorflow/tools/pip_package:build_pip_package
bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
pip3 install --user --ignore-installed /tmp/tensorflow_pkg/tensorflow-1.4.1-cp36-cp36m-linux_x86_64.whl

The commands are summarized from the Tensorflow installation guide (linked).

Post has attachment
I found this blog post on tuning PostgreSQL for a developer laptop very useful.

Really liked the style:
1) Has a clear target audience (MBP laptop, developer who needs to do some heavy data analysis)
2) Picks the most important knobs (subjective, of course), describes them, and shows good settings for MBP laptops.

Given Apple's lack of effort in upgrading MBPs (my laptops have had 16GB of RAM since 2012), the article has also aged extremely well. The only outdated bits are due to PostgreSQL upgrades -- wal_buffers can now be auto-set, and checkpoint_segments is no longer necessary (new setting has a better default).

Straightforward way to get going with Tensorflow and Jupyter on Fedora:

sudo dnf install python3-notebook python3-matplotlib
pip3 install --user --ignore-installed jupyter

Inside the notebook:
!pip3 install --user --ignore-installed tensorflow numpy matplotlib ipython

Post has attachment
C++ Deep learning library whose examples are smaller than other libraries written in higher-level languages. (ahem, TensorFlow) It's always nice to see small, straightforward APIs.
Wait while more posts are being loaded