Torchegranate

Latest version: v0.5.0

Safety actively analyzes 682229 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

0.1.2

=============

Highlights
----------

- Initial GPU support for all models and distributions
- Serialization support through native PyTorch

HiddenMarkovModel
-----------------

- Added labeled training by passing in an optional set of integer labels
alongside the data.
- Changed defaults from 64-bit arrays to 32-bit to potentially make use of
tensor cores when using a GPU.

0.1.1

=============

Highlights
----------

- Fixed a minor bug that surfaced after 0.1.0 release.

0.1.0

==============

Highlights
----------

- This is the first public release of torchegranate and so this entry will
document changes versus the original pomegranate repository.
- The biggest change is that the entire package has been rewritten in
PyTorch, with no Cython components, to installing and maintenance should
be easier.
- The `from_samples` method has removed in favor of the sklearn style
of not passing in parameters to the initialization and subsequently
calling `fit.`
- All inputs are meant to be 2D (or 3D for Markov chains and HMMs) numeric
values.
- Support for passing in symbols or defining distributions over arbitrary
symbols has been removed. Please manually encode your symbols or use the
appropriate sklearn function to do so.
- Models can be more tightly integrated with neural networks implemented
in PyTorch, e.g. a distribution can take the predicted outputs from a
model and use them to evaluate data.
- Sampling has not yet been added.


Distributions
-------------

- Distribution names now omit the word "Distribution", e.g.
"NormalDistribution" -> "Normal".
- "DiscreteDistribution" -> Categorical
- "JointProbabilityTable" -> JointCategorical
- "ConditionalProbabilityTable" -> ConditionalCategorical
- All distributions are now multivariate. Most are independent, in the sense
that each feature is evaluated separately. The log/probabilities returned
are evaluated under the entire feature set.
- Kernel densities have been removed for being annoying.
- Neural network wrappers have been removed due to the switch to PyTorch.
- ZeroInflated has been added as a wrapper for data that is a mixture of
zeroes and true signal
- StudentT has been added as a distribution
- DiracDelta has been added as a distribution


NaiveBayes.pyx
--------------

- The NaiveBayes model has been removed, with BayesClassifier containing
its entire functionality.


parallel.pyx
------------

- These parallel functions have been deprecated because of the switch to
PyTorch.


io.py
-----

- These functions have been deprecated because of the API switch.

Links

Releases

Has known vulnerabilities

© 2024 Safety CLI Cybersecurity Inc. All Rights Reserved.