Software

Parsita

Parsita is a parser combinator library for Python. It is hosted on Github and available on PyPI. Its main goal is a maximally simple interface for parsing. The API is inspired by Scala's parser combinator library, but the operators were carefully selected to have intuitive meaning and precedence in Python. A unique feature of Parsita is that it uses metaclass magic to allow for forward declarations of values by converting names not found during class body execution into forward declaration objects resolved during class instantiation.

Tabeline

Tabeline is a data table and data grammar library for Python. It is inspired by dplyr in R and uses Polars underneath. It provides and easy-to-use interface for doing data manipulations by providing the standard verbs (like select, filter, and mutate) as methods on its central DataTable class and having those methods accept strings (like table.filter("t < 24")) which are parsed and executed in the context of the table.

Tensora

Tensora is a sparse and dense tensor algebra library for Python. It is based on the Tensor Algebra Compiler (TACO). The user can create tensors whose formats are varying combinations of dense and sparse dimensions. The tensor can be populated from a variety of sources, such as dictionaries of keys, list of tuples, tuples of lists, NumPy arrays, and SciPy sparse matrices. While basic mathematical operators, such as __add__, __sub__, __matmul__, are defined, the most important feature is the evaluate function, which takes a string like "y(i) = A(i,j) * x(j)" and executes that expression on the given tensors. Under the hood, evaluate parses the tensor algebra expression, validates it, turns it into C code, invokes a C compiler on that code, executes the code on the given tensors, and returns the result to the user. All in all, it provides an easy-to-use interface to a very powerful tool.

Serialite

Serialite is serialization/deserialization library for Python. It is similar to Pydantic, but is much more strict. It's main abstract base class anticipates to_data and from_data methods. It provides the serializable decorator, which can be applied to data classes to generate those methods from the data class fields. It also provides the abstract_serializable decorator, which can be applied to sealed classes to generate those methods using a "_type" discriminator which switches on the __subclasses__ of the base class. Serialite implements enough of the Pydantic interface so that it can be used by FastAPI.