Skip to main content

Quickstart

From an approved subscription to your first query, in Python, using one day of matches. Getting the Data explains each step in more depth, and the CSDS Spec lists every channel and column.

1. Subscribe​

Subscribe on the product page. We review each request, which takes a few days; say what you plan to do with the data.

Once you are approved, open AWS Data Exchange in the us-east-1 region, choose Entitled data, open the product, then the data set pureskillgg-csgo-production-dataexchange-csds-0, and copy its Data set ID from the data set overview.

2. Install​

You need Python 3.11 or newer, AWS credentials that boto3 can find, the AWS CLI for step 4, and the two PureSkill.gg libraries:

pip install "pureskillgg-dsdk>=4.0.1" "pureskillgg-csgo-dsdk>=3.3.1"

The data set lives in us-east-1, so point boto3 there, for example with the environment variable AWS_DEFAULT_REGION=us-east-1.

3. Export one day to S3​

Data Exchange delivers a revision into an S3 bucket you own. Create one in us-east-1 first, so the export doesn't add data transfer charges, then export one day's revision:

from datetime import date, timedelta

from pureskillgg_dsdk import export_multiple_adx_dataset_revisions_to_s3

DATA_SET_ID = "paste the Data set ID here"
DAY = date.today() - timedelta(days=7)  # any day from the past year

export_multiple_adx_dataset_revisions_to_s3(
    "your-bucket",
    DATA_SET_ID,
    start_date=DAY.isoformat(),
    end_date=(DAY + timedelta(days=1)).isoformat(),
)
print("exported", DAY.strftime("csds/%Y/%m/%d"))

That exports the revision created on DAY: every match processed that day, usually 60 to 150 matches of 43 objects each. Revisions are pruned once they are about a year old, so pick a day from the past year. The Cost FAQ says what that costs.

4. Download it​

Copy the day to your machine with the AWS CLI, putting in the csds/YYYY/MM/DD path that step 3 printed. Keep that part of the path: the files' keys are how the libraries find them.

aws s3 sync s3://your-bucket/csds/YYYY/MM/DD/ data/csds/YYYY/MM/DD/ --exclude "*/player_vector" --exclude "*/player_status"

The two excluded channels are the per-tick telemetry, about 7 MB of each match's 10 MB. Leave the --exclude options out if you want player positions every tick.

5. Load one match​

Each match is a folder holding one file per channel and an index object named csds that lists them. Point a loader at a match's index:

import glob
import os

from pureskillgg_dsdk import DsReaderFs, GameDsLoader

ROOT = "data"
keys = sorted(
    os.path.relpath(path, ROOT).replace(os.sep, "/")
    for path in glob.glob(os.path.join(ROOT, "csds", "*", "*", "*", "*", "csds"))
)
print(len(keys), "matches")

loader = GameDsLoader(reader=DsReaderFs(root_path=ROOT, manifest_key=keys[0]))
match = loader.get_channels([{"channel": "header"}, {"channel": "player_death"}])
print(match["header"][["map_name", "match_date", "platform"]])
print(match["player_death"][["round", "tick", "weapon_name", "is_headshot"]].head())

If you downloaded player_vector, add the ten columns it computes on load, such as velocity and movement angle:

from pureskillgg_csgo_dsdk import add_player_vector_derived_columns

player_vector = loader.get_channel({"channel": "player_vector"})
add_player_vector_derived_columns(player_vector)

6. Query the whole day​

A tome is one table of a channel across many matches, with each row tagged with its match's match_key. Build tomes for the day, then query them with pandas:

from pureskillgg_dsdk import TomeCuratorFs

day = "-".join(keys[0].split("/")[1:4])  # the day you downloaded, from its key
curator = TomeCuratorFs(
    default_header_name=f"header.{day},{day}",
    ds_type="csds",
    tome_collection_root_path="tomes",
    ds_collection_root_path=ROOT,
)
tomes = curator.build_basic_tomes(["player_death"])
header = tomes.header.get_dataframe()
deaths = tomes.tomes["player_death"].get_dataframe()

kills = deaths[
    deaths["attacker_id"].notna() & (deaths["attacker_id"] != deaths["player_id"])
]
kills = kills.merge(header[["key", "map_name"]], left_on="match_key", right_on="key")
print(kills.groupby(["map_name", "weapon_name"]).size().sort_values().tail(10))

That keeps the deaths another player caused, dropping falls, suicides and the like, and prints the ten most common map and weapon pairs among the day's kills.

Next​