PECOS - Predictions for Enormous and Correlated Output Spaces

PECOS is a versatile and modular machine learning (ML) framework for fast learning and inference on problems with large output spaces, such as extreme multi-label ranking (XMR) and large-scale retrieval. PECOS' design is intentionally agnostic to the specific nature of the inputs and outputs as it is envisioned to be a general-purpose framework for multiple distinct applications.

Given an input, PECOS identifies a small set (10-100) of relevant outputs from amongst an extremely large (~100MM) candidate set and ranks these outputs in terms of relevance.

Features

Extreme Multi-label Ranking and Classification

X-Linear (pecos.xmc.xlinear): recursive linear models learning to traverse an input from the root of a hierarchical label tree to a few leaf node clusters, and return top-k relevant labels within the clusters as predictions. See more details in the PECOS paper (Yu et al., 2020).
- fast real-time inference in C++
- can handle 100MM output space
XR-Transformer (pecos.xmc.xtransformer): Transformer based XMC framework that fine-tunes pre-trained transformers recursively on multi-resolution objectives. It can be used to generate top-k relevant labels for a given instance or simply as a fine-tuning engine for task aware embeddings. See technical details in XR-Transformer paper (Zhang et al., 2021).
- easy to extend with many pre-trained Transformer models from huggingface transformers.
- establishes the State-of-the-art on public XMC benchmarks.
ANN Search with HNSW (pecos.ann.hnsw): a PECOS Approximated Nearest Neighbor (ANN) search module that implements the Hierarchical Navigable Small World Graphs (HNSW) algorithm (Malkov et al., TPAMI 2018).
- Supports both sparse and dense input features
- SIMD optimization for both dense/sparse distance computation
- Supports thread-safe graph construction in parallel on multi-core shared memory machines
- Supports thread-safe Searchers to do inference in parallel, which reduces inference overhead

Requirements and Installation

Python (3.7, 3.8, 3.9, 3.10)
Pip (>=19.3)

See other dependencies in setup.py You should install PECOS in a virtual environment. If you're unfamiliar with Python virtual environments, check out the user guide.

Supporting Platforms

Ubuntu 20.04 and 22.04
Amazon Linux 2

Installation from Wheel

PECOS can be installed using pip as follows:

python3 -m pip install libpecos

Installation from Source

Prerequisite builder tools

For Ubuntu (20.04, 22.04):

sudo apt-get update && sudo apt-get install -y build-essential git python3 python3-distutils python3-venv

For Amazon Linux 2:

sudo yum -y install python3 python3-devel python3-distutils python3-venv && sudo yum -y groupinstall 'Development Tools'

One needs to install at least one BLAS library to compile PECOS, e.g. OpenBLAS:

For Ubuntu (20.04, 22.04):
```
sudo apt-get install -y libopenblas-dev
```

For Amazon Linux 2:

sudo amazon-linux-extras install epel -y
sudo yum install openblas-devel -y

Install and develop locally

git clone https://github.com/amzn/pecos
cd pecos
python3 -m pip install --editable ./

Quick Tour

To have a glimpse of how PECOS works, here is a quick tour of using PECOS API for the XMR problem.

Toy Example

The eXtreme Multi-label Ranking (XMR) problem is defined by two matrices

instance-to-feature matrix X, of shape N by D in SciPy CSR format
instance-to-label matrix Y, of shape N by L in SciPy CSR format

Some toy data matrices are available in the tst-data folder.

PECOS constructs a hierarchical label tree and learns linear models recursively (e.g., XR-Linear):

>>> from pecos.xmc.xlinear.model import XLinearModel
>>> from pecos.xmc import Indexer, LabelEmbeddingFactory

# Build hierarchical label tree and train a XR-Linear model
>>> label_feat = LabelEmbeddingFactory.create(Y, X)
>>> cluster_chain = Indexer.gen(label_feat)
>>> model = XLinearModel.train(X, Y, C=cluster_chain)
>>> model.save("./save-models")

After learning the model, we do prediction and evaluation

>>> from pecos.utils import smat_util
>>> Yt_pred = model.predict(Xt)
# print precision and recall at k=10
>>> print(smat_util.Metrics.generate(Yt, Yt_pred))

PECOS also offers optimized C++ implementation for fast real-time inference

>>> model = XLinearModel.load("./save-models", is_predict_only=True)
>>> for i in range(X_tst.shape[0]):
>>>   y_tst_pred = model.predict(X_tst[i], threads=1)

Citation

If you find PECOS useful, please consider citing the following paper:

PECOS: Prediction for Enormous and Correlated Output Spaces (Yu et al., JMLR 2022) [bib]

Some papers from PECOS team:

License

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

amzn / pecos

readme