manifoldpy

This repo, in particular the API, is no longer being maintained. Manifold now releases full data dumps, which are much easier to work with than the API. I am still happy to review PRs.

Tools for analyzing Manifold Markets data. Currently has bindings for their API, and code for computing various accuracy metrics (Brier score, log score, calibration).

Full Documentation.

Calibration

This is the most recent calibration graph for Manifold Markets. Markets with less than 4 unique traders are excluded. The calibration is generated from the probability halfway between market open and market close.

Installation

The package is on PyPI as manifoldpy. It can be installed with:

pip install manifoldpy

However, you may want to install the most up-to-date version from github instead if you encounter any issues:

pip install --upgrade https://github.com/vluzko/manifoldpy/tarball/main

Basic Usage

Get a list containing every market:

from manifoldpy import api
markets = api.get_all_markets()

Manifold also has a POST API that lets you make/resolve/bet on markets. This requires you to have an API key (which you can generate on your Manifold profile page). Here's an example for making a bet on a binary market:

from manifoldpy import api
# Amount to spend
amount = 100
# The ID of the market to bet on
contract_id = "8Lt9ZTHCPCK58gtn0Y8n"
# The outcome you want to bet on.
outcome = "YES"
api.make_bet(YOUR_API_KEY, amount, contract_id, outcome)

If you want to avoid passing the key every time, you can instead create an APIWrapper and use that:

from manifoldpy import api
wrapper = api.APIWrapper(YOUR_API_KEY)
amount = 100
contract_id = "8Lt9ZTHCPCK58gtn0Y8n"
outcome = "YES"
wrapper.make_bet(amount, contract_id, outcome)

Get a market's history of probabilities:

from manifoldpy import api
market = api.get_market("market_id")
times, probabilities = market.probability_history()

Generate a basic calibration graph:

from manifoldpy import api, calibration
full_markets = api.get_full_markets()
binary = [
    m for m in full_markets if isinstance(m, api.BinaryMarket) and m.isResolved
]
df, histories = calibration.build_dataframe(binary)

yes_markets = df[df["resolution"] == "YES"]
no_markets = df[df["resolution"] == "NO"]

# Calibration at start
yes_probs = yes_markets["start"]
no_probs = no_markets["start"]
calibration = calibration.market_set_accuracy(yes_probs, no_probs)

There are additional examples for analyzing market calibration in the scripts/ directory.