Estimation of multinomial logit models in python
The following example analyzes choices of fishing modes. See the data here and more information about the data here. The parameters are:
X: Data matrix in long format (numpy array, shape [n_samples, n_fvariables])y: Binary vector of choices (numpy array, shape [n_samples, ])varnames: List of variable names. Its length must match number of columns inXalternatives: List of alternatives names or codes.asvars: List of alternative specific variablesisvars: List of individual specific variables The current version ofpymlogitonly supports data in long format.
from pymlogit.linear import MultinomialModel
import pandas as pd
df = pd.read_csv("examples/data/fishing_long.csv")
varnames = ['income','price']
X = df[varnames].values
y = df['choice'].values
model = MultinomialModel()
model.fit(X,y,isvars = ['income'], asvars=['price'],alternatives=['beach','boat','charter','pier'],varnames= varnames)
model.summary()Optimization succesfully completed after 11 iterations.
-----------------------------------------------------------------------------------------
Coefficient Estimate Std. Error z-value Pr(>|z|)
-----------------------------------------------------------------------------------------
_intercept.boat 0.4928935957 0.2053370982 2.4004118111 0.0449401617 .
_intercept.charter 1.8540668405 0.2097451458 8.8396173995 0.0000000000 ***
_intercept.pier 0.7526662342 0.2042533633 3.6849637242 0.0009279897 **
income.boat 0.0000933295 0.0000471101 1.9810953073 0.1122778277
income.charter -0.0000324867 0.0000478462 -0.6789828929 0.6333945307
income.pier -0.0001267191 0.0000465724 -2.7209078124 0.0198559789 .
price -0.0255642838 0.0015153615 -16.8700891463 0.0000000000 ***
-----------------------------------------------------------------------------------------
Significance: *** 0 ** 0.001 * 0.01 . 0.05
Log-Likelihood= -1220.535
Install using pip:
pip install pymlogit
Alternatively, you can download source code and import pymlogit.linear.MultinomialModel
The current version allows estimation of:
- Multinomial Logit Models: Models with individual specific variables
- Conditional Logit Models: Models with alternative specific variables
- Models with both, individual and alternative specific variables
The current version does not support models with panel data.