Skip to content

Latest commit

 

History

History
189 lines (139 loc) · 6.63 KB

README.md

File metadata and controls

189 lines (139 loc) · 6.63 KB
Raw

daSC2 - Data Analytics for StarCraft II

daSC2 is a data analytics toolset under development for StarCraft 2 by David Kuhta and Anshul Sacheti.

Quick Start Guide

Get daSC2

PyPI

The easiest way to get daSC2 is to use pip:

$ pip install dasc2

or if you desire a visual progress bar:

$ pip install dasc2[with-progress-bar]

That will install the dasc2 package along with all the required dependencies.

Pip will install a few of the binaries to your bin directory.

Git

Alternatively you can install daSC2 with git.

  1. Clone the daSC2 repo
  2. Install the dependencies and dasc2 package:
$ git clone https://github.com/davidkuhta/dasc2.git
$ pip install dasc2/

Available Commands

Download replays

This tool allows for the downloading of .SC2Replay archives via Blizzard's API.

You can download replays using the command: dasc2_download

Required Flags

flag description
--key Blizzard API client key
--secret Blizzard API client secret
--version Desire StarCraft 2 Build Version

Optional Flags

flag description default
-h, --help Show help message
--archives_dir Directory in which to save archives ./archives

Extract Replays

This tool simplifies extracting of the StarCraft replay files from the previously download archives.

You can extract replays using the command: dasc2_extract

In accordance with Blizzard's EULA you will be prompted for agreement to the Blizzard AI and Machine Learning License

Optional Flags

flag description default
-h, --help Show help message
--archives_dir Directory in which archives are located ./archives
--replays_dir Directory in which to extract replays ./replays
--remove_archives Flag denoted to remove the archives upon extraction

Filter replays

This tool filters .SC2Replay files within the denoted directory, generating a filter file containing information related to the replay files meeting the filtering criteria:

You can filter replays using the command: dasc2_filter

Optional Flags

flag description default
-h, --help Show help message
--replays_dir Select replay directory ./replays
--filters_dir Select directory to place filters ./filters
--min_mmr Set the minimum player MMR 1000
--min_apm Set the minimum player APM 10
--races Filter on certain races Zerg Terr Prot
--winning_races Filter on whether the winner belongs to a certain race Zerg Terr Prot
--game_map Filter based on map title all maps
--build Filter based on game build version
--full_path Flag denoting filters should list full path of replays

Note: no commas needed between race names

Generate replay states

This tool generates replay states for a list of replays denoted in a filter file. Note corrupted replays will be skipped so yield may be less than 100% as compared to the input list.

You can generate replay states using the command: dasc2_states

Required Flags

flag description
--replay_list File containing list of replays to evaluate
--filter_file File cotaining filter information

Optional Flags

flag description default
-h, --help Show help message
--parallel How many instances to run in parallel 1
--step_mul How many game steps per observation 1
--states_dir Directory to save states ./states
--print_time Interval between stat prints and data saves in seconds 10
--winner_only Generate states for winner only False

Build Orders

This tool generates json files containing replay information including build orders for both a player and their opponent

You can generate replay states using the command: dasc2_build_orders

Required Flags

flag description
--replay_list File containing list of replays to evaluate

Optional Flags

flag description default
-h, --help Show help message
--states_dir Directory where states are located ./states
--build_orders_dir Directory in which to generate build orders ./build_orders

Template Agent

This tool generates a duplicate of existing agents for use as templates.

Available agents include:

  • dasc2 (An RL agent)
  • base (pysc2's base agent)
  • scripted (pysc2's scripted agent)
  • random (pysc2's random agent)

You can generate replay states using the command: dasc2_template_agents

Required Flags

flag description
--agent Agent to create a template of.

Optional Flags

flag description default
-h, --help Show help message
--agents_dir Directory where agents are to be generated ./agents
--name The filename to rename the agent to ./build_orders

Replays

Obtaining Replays

  1. dasc2_download
  2. You can evaluate your own replays
  3. Blizzard
    1. Download replay packs linked on the s2client-proto repo

Attribution

We'd like to thank Blizzard, Deepmind, Arthur Juliani, and Cole Maclean, for the influence each of their projects had on daSC2. Any code utilized was in accordance with it's respective license.