2.2.1

• Fix issue where bootstrap_stats would always run parametric bootstrapping (#57 )
• Expose the missing parametric argument in bootstrap_stats

2.2.0

• StepMixClassifier object for supervised learning (binary and multiclass classification).
• Parametric Bootstrapping.
• BLRT test for selecting the number of components (#46).
• Sampling categorical emission models now returns integer-encoded categories.

2.1.3

• Improve example scripts
• Fix version issue. 2.1.2 did not include some changes on master

2.1.1

• Minor edits to docstring and READMEs.

2.1.0

• Improvements to minimize the use of pivot_table for users:
  • New methods to directly access measurement and structural parameters dataframes with latent class as columns
  • New bootstrap_stats method that returns both bootstrap samples and some parameter statistics (mean and standard deviation) as dataframes
• verbose=1 now prints parameters using class as columns

2.0.0

• StepMix and Emission models can now return parameters as dataframes.
• StepMix remembers variables names when input is a DataFrame. This is reflected in the logs.
• Bootstrap API overhaul.
  • Bootstrap is now a StepMix method.
  • Bootstrap no longer fits the main estimator (#36).
  • Bootstrap returns parameters as a long-form dataframe.
  • Remove custom plotting functions. Users can now easily use bootstrap output with seaborn.
• CAIC and SABIC (#46).
• Fix bug when sampling a StepMix model with only a measurement model (#45).
• Update docstring of X and Y matrices (#37).

1.2.5

• Always maximize joint likelihood in the first step of 3-step estimation (#43 )

1.2.4

• Fix issues with parameter counting in nested models (#41)

1.2.3

• Fix wrong parameter count in Gaussian models (#40)
• Fix wrong parameter count in covariate model. Do not include class weights, since they are not part of the model
• Fix doc building issues related to recent sklearn changes to BaseEstimator
• Better Quickstart example in README and doc

1.2.1

• Fix error occurring when some stepwise estimators were using Dataframes as input (#39)
• Clip probabilities in entropy computations to avoid log 0 errors