carball

Carball is an open-source project that combines multiple tools for decompiling Rocket League replays and then analysing them.

Requirements

• Python 3.6.7+ (3.7 and 3.8 included)
• Windows, Mac or Linux

Install

Install from pip:

pip install carball

Clone for development

Windows

git clone https://github.com/SaltieRL/carball
cd carball/
python init.py

Linux

git clone https://github.com/SaltieRL/carball
cd carball/
./_travis/install-protoc.sh
python init.py

Mac

In MacOS Catalina, zsh replaced bash as the default shell, which may cause permission issues when trying to run install-protoc.sh in the above fashion. Simply invoking bash should resolve this issue, like so:

git clone https://github.com/SaltieRL/carball
cd carball/
bash ./_travis/install-protoc.sh
python init.py

Apple's decision to replace bash as the default shell may foreshadow the removal of bash in a future version of MacOS. In such a case, Homebrew users can install protoc by replacing bash ./travis/install-protoc.sh with brew install protobuf.

Examples / Usage

One of the main data structures used in carball is the pandas.DataFrame, to learn more, see its wiki page.

Decompile and analyze a replay:

import carball

analysis_manager = carball.analyze_replay_file('9EB5E5814D73F55B51A1BD9664D4CBF3.replay', 
                                      output_path='9EB5E5814D73F55B51A1BD9664D4CBF3.json', 
                                      overwrite=True)
proto_game = analysis_manager.get_protobuf_data()

# you can see more example of using the analysis manager below

Just decompile a replay to a JSON object:

import carball

_json = carball.decompile_replay('9EB5E5814D73F55B51A1BD9664D4CBF3.replay', 
                                output_path='9EB5E5814D73F55B51A1BD9664D4CBF3.json', 
                                overwrite=True)

Analyze a JSON game object:

import carball
import gzip
from carball.json_parser.game import Game
from carball.analysis.analysis_manager import AnalysisManager

# _json is a JSON game object (from decompile_replay)
game = Game()
game.initialize(loaded_json=_json)

analysis_manager = AnalysisManager(game)
analysis_manager.create_analysis()
    
# return the proto object in python
proto_object = analysis_manager.get_protobuf_data()

# return the proto object as a json object
json_oject = analysis_manager.get_json_data()

# return the pandas data frame in python
dataframe = analysis_manager.get_data_frame()

You may want to save carball analysis results for later use:

# write proto out to a file
# read api/*.proto for info on the object properties
with open('output.pts', 'wb') as fo:
    analysis_manager.write_proto_out_to_file(fo)
    
# write pandas dataframe out as a gzipped numpy array
with gzip.open('output.gzip', 'wb') as fo:
    analysis_manager.write_pandas_out_to_file(fo)

Read the saved analysis files:

import gzip
from carball.analysis.utils.pandas_manager import PandasManager
from carball.analysis.utils.proto_manager import ProtobufManager

# read proto from file
with open('output.pts', 'rb') as f:
    proto_object = ProtobufManager.read_proto_out_from_file(f)

# read pandas dataframe from gzipped numpy array file
with gzip.open('output.gzip', 'rb') as f:
    dataframe = PandasManager.read_numpy_from_memory(f)

Command Line

Carball comes with a command line tool to analyze replays. To use carball from the command line:

carball -i 9EB5E5814D73F55B51A1BD9664D4CBF3.replay --json analysis.json

To get the analysis in both json and protobuf and also the compressed replay frame data frame:

carball -i 9EB5E5814D73F55B51A1BD9664D4CBF3.replay --json analysis.json --proto analysis.pts --gzip frames.gzip

Command Line Arguments

usage: carball [-h] -i INPUT [--proto PROTO] [--json JSON] [--gzip GZIP] [-sd]
               [-v] [-s]

Rocket League replay parsing and analysis.

optional arguments:
  -h, --help            show this help message and exit
  -i INPUT, --input INPUT
                        Path to replay file that will be analyzed. Carball
                        expects a raw replay file unless --skip-decompile is
                        provided.
  --proto PROTO         The result of the analysis will be saved to this file
                        in protocol buffers format.
  --json JSON           The result of the analysis will be saved to this file
                        in json file format.
  --gzip GZIP           The pandas dataframe will be saved to this file in a
                        compressed gzip format.
  -v, --verbose         Set the logging level to INFO. To set the logging
                        level to DEBUG use -vv.
  -s, --silent          Disable logging altogether.

Pipeline

If you want to add a new stat it is best to do it in the advanced stats section of the pipeline. You should look at:

Stat base classes

Where you add a new stat

If you want to see the output format of the stats created you can look here

Compile the proto files by running in this directory setup.bat (Windows) or setup.sh (Linux/mac)

Tips

Linux set python3.6 as python:

sudo update-alternatives --install /usr/bin/python python /usr/bin/python3.6 1

This assumes you already have 3.6 installed.

Developing

Everyone is welcome to join the carball (and calculated.gg) project! Even if you are a beginner, this can be used as an opportunity to learn more - you just need to be willing to learn and contribute.

Usage of GitHub

All contributions end up on the carball repository. If you are new to the project you are required to use your own fork for first changes. If you do not have any previous git / github experience that is completely fine - we can help with it. If we believe that you are comitted to working on the project and have experience in git we may give you write access so that you no longer have to use a fork. Nonetheless, please wait until your contrubtion is ready for a review to make the pull request because that will save resources for our tests and reduce spam. For testing you should use your own fork, but take note that some carball tests may fail on a fork

Learning about carball

Currently, there is active creation of the carball wiki on GitHub - it aims to provide all relevant information about carball and development, so if you are a beginner, definitely have a look there. If you can't find information that you were looking for, your next option is the calculated.gg Discord server, where you may send a message to the #help channel.

The carball code is also documented, although sparsely. However, you still may find information there, too.

Testing

The main requirement is to run PyTest. If you are using an IDE that supports integrated testing (e.g. PyCharm), you should enable PyTest there. The secondary requirement (to compile the proto files) is to run the appropriate setup file (setup.bat for Windows, setup.sh for Linux/Mac).

If you've never tested your code before, it is a good idea to learn that skill with PyTest! Have a look at their official documentation, or any other tutorials.

carball Performance

Carball powers calculated.gg, which analyses tens of thousands of replays per day. Therefore, performance is very important, and it is monitored and controlled using PyTest-Benchmarking, which is implemented via GitHub Actions. However, you may see your contribution's performance locally - look into PyTest-Benchmarking documentation. If your contribution is very inefficient - it will fail automatically.

If you wish to see the current carball analysis performance, it is split into 5 replay categories, and can be accessed below:

• Short Sample
  • A very short soccar replay - for fast benchmarking.
• Short Dropshot
  • A very short dropshot replay - to test dropshot performance.
• Rumble
  • A full game of rumble - to test rumble performance.
• RLCS
  • A full soccar RLCS game.
• RLCS (Intensive)
  • A full soccar RLCS game, but run with the intense analysis flag.