trader data
Hi Zhengjie,
Great work on making these data available!
Is the quant.parquet and users.parquet maintained on hugging face generated from the same trades.parquet file? Or they are from different snapsot of trade.parquet. As I found some trades does not show up in users.parquet.
Thanks!
Hi，
The address 0x7c3db723f1d4d8cb9c550095203b686cb11e5c6b is indeed present in the dataset:
As maker: 157,911 trades
As taker: 103,733 trades
Total: 261,644 trades
The maker and taker columns store raw wallet addresses. You can search like this:
import pyarrow.parquet as pq
import pyarrow.compute as pc
path = "trades.parquet"
address = "0x7c3db723f1d4d8cb9c550095203b686cb11e5c6b"
table = pq.read_table(path, columns=["maker", "taker"])
maker_count = pc.sum(pc.equal(pc.utf8_lower(table["maker"]), address)).as_py()
taker_count = pc.sum(pc.equal(pc.utf8_lower(table["taker"]), address)).as_py()
print(f"maker: {maker_count}, taker: {taker_count}")
We recommend using pyarrow directly rather than pandas for better memory efficiency on this large dataset.
Regarding your second question: trades.parquet and quant.parquet were updated in early March with the latest data, but users.parquet was not — it reflects an older snapshot. We have a continuously running data pipeline, but due to storage constraints and the fact that user-level aggregations are not central to our current research, we have not been updating users.parquet.
If you need up-to-date user-level data, you can derive it from trades.parquet directly. If you're interested in collaborating or need further assistance, feel free to reach out!
Thank you! It turns out that some addresses in the trader dataset were capitalized. After converting the addresses to lowercase, I was able to find this trader.