Numerical python

Warning: python precompiled binaries are slow and horrible

Specific CPU and GPU support in python precompiled binaries is weak and outdated. For maximum performance we should be compiling them.

zarr

Zarr is a format for the storage of chunked, compressed, N-dimensional arrays. …

Highlights:

• Create N-dimensional arrays with any NumPy dtype.
• Chunk arrays along any dimension.
• Compress and/or filter chunks using any NumCodecs codec.
• Store arrays in memory, on disk, inside a Zip file, on S3, …
• Read an array concurrently from multiple threads or processes.
• Write to an array concurrently from multiple threads or processes.
• Organize arrays into hierarchies via groups.

Resembles HDF5 but makes fewer assumptions about the storage backend and more assumptions about the language frontend.

Dask

dask is a flexible library for parallel computing in Python.

See the howto for some examples of use cases.

Dask is composed of two parts:

1. Dynamic task scheduling optimized for computation. This is similar to Airflow, Luigi, Celery, or Make, but optimized for interactive computational workloads.

2. “Big Data” collections like parallel arrays, dataframes, and lists that extend common interfaces like NumPy, Pandas, or Python iterators to larger-than-memory or distributed environments. These parallel collections run on top of dynamic task schedulers.

Dask emphasizes the following virtues:

• Familiar: Provides parallelized NumPy array and Pandas DataFrame objects

• Flexible: Provides a task scheduling interface for more custom workloads and integration with other projects.

• Native: Enables distributed computing in pure Python with access to the PyData stack.

• Fast: Operates with low overhead, low latency, and minimal serialization necessary for fast numerical algorithms

• Scales up: Runs resiliently on clusters with 1000s of cores

• Scales down: Trivial to set up and run on a laptop in a single process

• Responsive: Designed with interactive computing in mind, it provides rapid feedback and diagnostics to aid humans

HDF5

h5py provides a simple, easy and efficient interface to HDF5 files, which are an easy way of loading and saving arrays of numbers and indeed arbitrary data. I use this a lot but sometimes I get glitches on macos with passing around serialised H5py objects, which works fine on Linux.

xarray

xarray is a system which labels arrays, which is useful in keeping track of them for statistical analysis.

Xarray introduces labels in the form of dimensions, coordinates and attributes on top of raw NumPy-like multidimensional arrays, which allows for a more intuitive, more concise, and less error-prone developer experience.

This data model is borrowed from the netCDF file format, which also provides xarray with a natural and portable serialization format. NetCDF is very popular in the geosciences, and there are existing libraries for reading and writing netCDF in many programming languages, including Python.

This system seems to be a developing lingua franca for the differentiable learning frameworks.

Compiling

e.g. for performance or invoking external binaries.

See compiling python.

Displaying numbers legibly

Easy, but documentation is hard to find.

print("{:10.4f}".format(x))

np.set_printoptions(edgeitems=5,
  linewidth=85, precision=4,
  suppress=True, threshold=500)

np.set_printoptions(edgeitems=3, infstr='inf',
  linewidth=75, nanstr='nan', precision=8,
  suppress=False, threshold=1000, formatter=None)

Einstein operations

numpy.einsum:

The Einstein summation convention can be used to compute many multi-dimensional, linear algebraic array operations. einsum provides a succinct way of representing these.

A non-exhaustive list of these operations, which can be computed by einsum, is shown below along with examples:

• Trace of an array, numpy.trace.
• Return a diagonal, numpy.diag.
• Array axis summations, numpy.sum.
• Transpositions and permutations, numpy.transpose.
• Matrix multiplication and dot product, numpy.matmul numpy.dot.
• Vector inner and outer products, numpy.inner numpy.outer.
• Broadcasting, element-wise and scalar multiplication, numpy.multiply.
• Tensor contractions, numpy.tensordot.
• Chained array operations, in efficient calculation order, numpy.einsum_path.

The subscripts string is a comma-separated list of subscript labels, where each label refers to a dimension of the corresponding operand. Whenever a label is repeated it is summed, so np.einsum('i,i', a, b) is equivalent to np.inner(a,b). If a label appears only once, it is not summed, so np.einsum('i', a) produces a view of a with no changes. A further example np.einsum('ij,jk', a, b) describes traditional matrix multiplication and is equivalent to np.matmul(a,b). Repeated subscript labels in one operand take the diagonal. For example, np.einsum('ii', a) is equivalent to np.trace(a).

Also available in all NN frameworks, e.g. in pytorch as torch.einsum.

Here are some introductions to tool:

• Tim Rocktäschel
• Chris Lemke , einsum - an underestimated function
• Understanding einsum for Deep learning: implement a transformer with multi-head self-attention from scratch

Higher performance, opt_einsum makes sure that the result is easy for the computer and not only the human:

Optimized einsum can significantly reduce the overall execution time of einsum-like expressions by optimizing the expression’s contraction order and dispatching many operations to canonical BLAS, cuBLAS, or other specialized routines. Optimized einsum is agnostic to the backend and can handle NumPy, Dask, PyTorch, Tensorflow, CuPy, Sparse, Theano, JAX, and Autograd arrays as well as potentially any library which conforms to a standard API.

References