Rodrigo Kumpera
Rodrigo Kumpera
run: `example_gen --king cb --count 100` Then use log_parser to look into it. You'll notice not 100 rewards were generated. This doesn't happen with `--kind cb` for some reason.
I'm using this rl.net.cli command: ``` dotnet .\Rl.Net.Cli.dll perf --actions 50 --shared-features 3 --action-features 10 --data 1 --config client.json ``` With the following client.json: ``` { "ApplicationID": "cf8a74ffaa1f4d1a872d251909bfa77d", "observation.file.name": "C:\\joiner-data\\observations.fb",...
Out current build story on windows needs some love. The problem: We currently have two build scripts for rlclientlib: cmake and msbuild. The CI coverage of the msbuild setup is...
There's a serialization bug in our v2 code that writes the event_id twice in the payload.
parser.py can will now parse all preamble messages in a file. rl.net.cli was not disposing its live model, now it does. Introduced some basic logging to the batching process so...
The two large file improvements are: `--short` and `--summary`. `--short`: produce a one line message per event. `--summary`: print one line per checkpoint message.
This task lists all testing scenarios we need to cover when dealing with the new file format. - when using dedup, an example with two or more identical actions.
It's better when example_gen explicitly set the value so it's controlled by its PRNG. The goal is to enable us to generate slot_id based test examples and not just slot_index.
Schema v2 has indexed rewards and we need binding support for it.
Schema v2 led serialization to happen on the main thread, which has two undesirable consequences. First, it moves work to application threads and that will increase the latency of each...