Skip to main content

Functions to compute probabilistic relevance scores from PHOC embeddings

Project description


Build Status

PyTorch functions to compute meaningful probabilistic relevance scores from PHOC (Pyramid of Histograms of Characters) embeddings. Although they are called Pyramid of Histograms of Characters, in practice they are a Pyramid of Bag of Characters. At the end, each word is represented by a high-dimensional binary vector.

See the wiki for additional details.


The library provides two functions: cphoc and pphoc, which are similar to SciPy's cdist and pdist:

Both functions can operate with PHOC embeddings in the probability space (where each dimension is a real number in the range [0, 1]), or in the log-probability space (where each dimension is the logarithm of a probability). These are also sometimes refered to as the Real and Log semirings.

import torch
from prob_phoc import cphoc, pphoc

x = torch.Tensor(...)
y = torch.Tensor(...)

# Compute the log-relevance scores between all pairs of rows in x, y.
# Note: x and y must have the PHOC log-probabilities.
logprob = cphoc(x, y)

# This is equivalent to:
logprob = cphoc(x, y, method="sum_prod_log")

# If your matrices have probabilities instead of log-probabilities, use:
prob = cphoc(x, y, method="sum_prob_real")

# Compute the log-relevance scores between all pairs of distinct rows in x.
# Note: The output is a vector with N * (N - 1) / 2 elements.
logprob = pphoc(x)


The easiest way is to install the package from PyPI:

pip install prob-phoc

If you want to install the latest version from the repository, clone it and use the script to compile and install the library.

python install

You will need a C++11 compiler (tested with GCC 4.9). If you want to compile with CUDA support, you will also need to install the CUDA Toolkit (tested with versions 8.0, 9.0 and 10.0)

Tests and benchmarks

After the installation, you can run the tests to ensure that everything is working fine.

python -m prob_phoc.test

I have also some benchmarks to compare CPU vs. CUDA, for different matrix sizes and float precision. These take quite a long to run, so don't hold your breath.

python -m prob_phoc.benchmark

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

prob_phoc-0.2.0.tar.gz (10.2 kB view hashes)

Uploaded source

Built Distributions

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page