mossspider

Targeted maximum likelihood estimation for network-dependent data


Keywords
tmle, network, causal-inference, network-analysis, targeted-maximum-likelihood
License
MIT
Install
pip install mossspider==0.0.3

Documentation

mossspider

tests

MossSpider

MossSpider provides an implementation of the targeted maximum likelihood estimator for network-dependent data (network-TMLE) in Python. Currently mossspider supports estimation of the conditional network mean for stochastic policies.

mossspider get its name from the spruce-fir moss spider, a tarantula that is both the world's smallest tarantula and native to North Carolina.

Installation

Installing:

You can install via python -m pip install mossspider

Dependencies:

The dependencies are: numpy, scipy, statsmodels, networkx, matplotlib. Notice that NetworkX must be at least 2.0.0 to work properly.

Getting started

To demonstrate mossspider, below is a simple demonstration of calculating the mean for the following data.

from mossspider import NetworkTMLE
from mossspider.dgm import uniform_network, generate_observed

First, we will use some built-in data generating functions

graph = uniform_network(n=500, degree=[1, 4])
graph_observed = generate_observed(graph)

Now, we can use NetworkTMLE to estimate the causal conditional mean under a stochastic policy. Here, the stochastic policy sets everyone's probability of action A=1 to 0.65.

ntmle = NetworkTMLE(network=graph_observed,
                    exposure='A',  # Exposure in graph
                    outcome='Y',   # Outcome in graph
                    verbose=True)  # Print model summaries
ntmle.exposure_model(model="W + W_sum")
ntmle.exposure_map_model(model='A + W + W_sum',  # Parametric model
                         measure='sum',          # Summary measure for A^s
                         distribution='poisson') # Model distribution to use
ntmle.outcome_model(model='A + A_sum + W + W_sum')
ntmle.fit(p=0.65, samples=500)
ntmle.summary()

For full details on using mossspider, see the full documentation and worked examples available at MossSpider website.