train

Continuous training of model as data arrives.

Optimizing dependencies only: https://www.reddit.com/r/rust/comments/gvrgca/this_is_a_neat_trick_for_getting_good_runtime/

Profiling, get a flamegraph:

install perf

sudo apt-get install linux-tools-common linux-tools-generic linux-tools-uname -r

sudo apt-get remove linux-tools-common linux-tools-generic linux-tools-uname -r

install flamegraph

cargo install flamegraph

run executable with profiling

flamegraph --no-inline -o ignore/my_flamegraph.svg -- ../target/debug/train 1 flamegraph --no-inline -o ignore/my_flamegraph.svg -- ../target/release/train 20

to do profiling: sudo sysctl kernel.perf_event_paranoid=-1 sudo bash -c "echo 0 > /proc/sys/kernel/kptr_restrict"

when done: sudo sysctl kernel.perf_event_paranoid=4 sudo bash -c "echo 1 > /proc/sys/kernel/kptr_restrict"

cargo flamegraph --no-inline -o ignore/my_flamegraph.svg -- ../target/release/train