A tool that generates statistics from official SSL log files.
You need to install following dependencies:
- Go
# Build the binaries and install them in $GOPATH/bin
make install
The ssl-match-stats
command will generate the statistics into an intermediate data structure *.{json|bin}
from which they can be exported. This must be done per tournament and division.
The command takes a list of log files as input:
mkdir -p stats
ssl-match-stats -parallel 16 -targetDir stats *.log.gz
The generated statistics can be exported into a PostgreSQL database. This makes it easier to query and analyze the data.
The command requires some parameters:
- tournament: A unique name for the tournament of the log files, for example 'RoboCup2019'
- division: The division of the log files, one of: 'DivA', 'DivB', 'none'
- sqlDbSource: A connection string to the target database
ssl-match-stats-db \
-parallel=16 \
-sqlDbSource="postgres://ssl_match_stats:ssl_match_stats@localhost:5432/ssl_match_stats?sslmode=disable" \
-tournament=Test \
-division=DivA \
stats/*.bin
Use the provided compose setup to run the database and metabase locally.
docker compose up -d
The database connection string
is: postgres://ssl_match_stats:ssl_match_stats@localhost:5432/ssl_match_stats?sslmode=disable
The host the statistics one a server, following the below instructions.
See Setup for Match Stats DB for instructions on setting up the PostgreSQL database.
See Setup for Metabase for instructions on setting up Metabase, an open-source BI software.
If you want to evaluate the start time offset between scheduled match start time and actual start time,
you need to add the scheduled time into matches.start_time_planned
manually.
To generate the sources from the .proto
files, run make proto
.
This tool reads log files one by one with the ssl-match-stats
command and creates a protobuf file based
on ssl_match_stats.proto.
This structure contains a MatchStats
object per match (log file), which contains:
- Some meta data like name, duration, etc.
- A list of all game phases (which map roughly to referee commands)
TeamStats
for each team
The incentive is that the MatchStats
structure contains an improved representation of the referee messages
for further processing. The match is split into game phases, where a new game phase is started for each new command
(except the goal commands). Each game phase has metadata, entry and exit states and game events attached to it.
TeamStats
contain aggregated final counters and timers.
With this improved representation, the data can be exported for further analysis.