A family of exponential-dispersion probability distributions that includes several continuous, discrete, and compound Poisson-gamma cases.
The Tweedie distribution is a family of probability distributions used as a candidate loss for count data. It is especially useful when outcomes may be zero yet otherwise continuous. section 5.3.1
The family includes:
Continuous normal, gamma, and inverse Gaussian distributions. section 5.3.1
The purely discrete scaled Poisson distribution. section 5.3.1
Compound Poisson–gamma distributions, with positive mass at zero and continuous positive values otherwise. section 5.3.1
Retail forecasting
Treat Tweedie loss as a second-line experiment alongside quantile losses when modeling count data for retail decisions. section 5.3.1
How Machine Learning System Design covers it
How deep it goes
Defines the family through its distributional members and zero-continuous behavior without deriving its mathematics.
Examples
Uses normal, gamma, inverse Gaussian, scaled Poisson, and compound Poisson–gamma distributions as family cases.
How it connects
Connects probability distributions with loss functions, count data, quantile loss, and compound Poisson–gamma models.
Ensemble Methods for Machine LearningA Tweedie distribution is a family of exponential-dispersion distributions that, for 1<p<2, represents compound Poisson-gamma behavior with a mass at zero and positive continuous values.
TweedieRegressorused by · TweedieRegressor is a scikit-learn estimator that fits generalized linear regression models with a Tweedie distribution and configurable power parameter.
LightGBMused by · LightGBM is an open-source gradient-boosting framework for efficiently training tree-based machine-learning models.
Poisson-gamma distributionuses · A Poisson-gamma distribution is a hybrid distribution combining a Poisson component for a point mass at zero with a gamma component for positive continuous values.
count dataaddresses · Count data consist of nonnegative integer-valued observations representing numbers of events, items, or occurrences.
Poisson distributionhas part · The Poisson distribution is a discrete probability distribution that models the count of events occurring independently at a constant average rate within a fixed interval or exposure.
gamma distributionhas part · The gamma distribution is a continuous probability distribution over positive values, characterized by shape and scale or rate parameters.
This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.
