ml4ir: Machine Learning Library for Information Retrieval

Setup

Requirements

• python3.6+
• pip3
• docker (version 18.09+ tested)

Using PIP

ml4ir can be installed as a pip package by using the following command

pip install  'git+https://git@github.com/salesforce/ml4ir#egg=ml4ir&subdirectory=python'

This will install ml4ir-0.0.1 (the current version). In future, when this package is available on PyPI, it will be as simple as pip install ml4ir

Docker (Recommended)

We have set up a docker-compose.yml file for building and using docker containers to train models.

To run unit tests

docker-compose up

To invoke ml4ir with custom arguments with docker, run

/bin/bash tools/run_docker.sh ml4ir \
	python3 ml4ir/base/pipeline.py
    <args>

For ranking applications, specifically, use

/bin/bash tools/run_docker.sh ml4ir \
	python3 ml4ir/applications/ranking/pipeline.py
    <args>

Refer to usage section below for details on how to run ml4ir - ranking

Check ml4ir/applications/ranking/scripts/example_run.sh for a predefined example run.

To run example invocation of ranking application with docker,

/bin/bash python/ml4ir/applications/ranking/scripts/example_run.sh

Virtual Environment

Install virtualenv

pip3 install virtualenv

Create new python3 virtual environment inside your git repo (it's .gitignored, don't worry)

cd $PLACE_YOU_CAlLED_GIT_CLONE/ml4ir
python3 -m venv python/env/.ml4ir_venv3

Activate virtualenv

cd python/
source env/.ml4ir_venv3/bin/activate

Install all dependencies (carefully)

pip3 install --upgrade setuptools
pip install --upgrade pip
pip3 install -r requirements.txt

Note, there are some AWS incompatibilities, gotta fix that, but you can ignore them for now

ERROR: botocore 1.14.9 has requirement docutils<0.16,>=0.10, but you'll have docutils 0.16 which is incompatible.
ERROR: awscli 1.17.9 has requirement docutils<0.16,>=0.10, but you'll have docutils 0.16 which is incompatible.
ERROR: awscli 1.17.9 has requirement rsa<=3.5.0,>=3.1.2, but you'll have rsa 4.0 which is incompatible.
ERROR: tensorflow-probability 0.8.0 has requirement cloudpickle==1.1.1, but you'll have cloudpickle 1.2.2 which is incompatible.
ERROR: apache-beam 2.18.0 has requirement dill<0.3.2,>=0.3.1.1, but you'll have dill 0.3.0 which is incompatible.
ERROR: apache-beam 2.18.0 has requirement httplib2<=0.12.0,>=0.8, but you'll have httplib2 0.17.0 which is incompatible.
ERROR: apache-beam 2.18.0 has requirement pyarrow<0.16.0,>=0.15.1; python_version >= "3.0" or platform_system != "Windows", but you'll have pyarrow 0.14.1 which is incompatible.
ERROR: tfx-bsl 0.15.3 has requirement absl-py<0.9,>=0.7, but you'll have absl-py 0.9.0 which is incompatible.
ERROR: tfx-bsl 0.15.3 has requirement apache-beam[gcp]<2.17,>=2.16, but you'll have apache-beam 2.18.0 which is incompatible.
ERROR: tensorflow-transform 0.15.0 has requirement absl-py<0.9,>=0.7, but you'll have absl-py 0.9.0 which is incompatible.

Note that pre-commit-hooks are required, and installed as a requirement if needed. If an error results that they didn't install, execute pre-commit install to install git hooks in your .git/ directory.

Set the PYTHONPATH environment variable

export PYTHONPATH=$PYTHONPATH:`pwd`/python

Usage

The entrypoint into the training or evaluation functionality of ml4ir is through ml4ir/base/pipeline.py and for application specific overrides, look at `ml4ir/applications/<eg: ranking>/pipeline.py

ml4ir Library

To use ml4ir as a deep learning library to build relevance models, look at the walkthrough under notebooks/PointwiseRankingDemo.ipynb or notebooks/PointwiseRankingDemo.html(contains architecture diagrams). The notebook walks one through building, training, saving, and the entire life cycle of a RelevanceModel from the bottom up. Additionally, the HTML version also sheds light on the design of ml4ir and the data format used.

Applications - Ranking

Examples

Using TFRecord

python ml4ir/applications/ranking/pipeline.py \
--data_dir ml4ir/applications/ranking/tests/data/tfrecord \
--feature_config ml4ir/applications/ranking/tests/data/config/feature_config.yaml \
--run_id test \
--data_format tfrecord \
--execution_mode train_inference_evaluate

Using CSV

python ml4ir/applications/ranking/pipeline.py \
--data_dir ml4ir/applications/ranking/tests/data/csv \
--feature_config ml4ir/applications/ranking/tests/data/config/feature_config.yaml \
--run_id test \
--data_format csv \
--execution_mode train_inference_evaluate

Running in inference mode using the default serving signature

python ml4ir/applications/ranking/pipeline.py \
--data_dir ml4ir/applications/ranking/tests/data/tfrecord \
--feature_config ml4ir/applications/ranking/tests/data/config/feature_config.yaml \
--run_id test \
--data_format tfrecord \
--model_file `pwd`/models/test/final/default \
--execution_mode inference_only

NOTE: Make sure to add the right data and feature config before training models.
TODO: describe how to do this

Running Tests

To run all the python based tests under ml4ir

python3 -m pytest

To run specific tests,

python3 -m pytest /path/to/test/module

Project Organization

The following structure is a little out of date (TODO(jake) - fix it!)

├── LICENSE
├── Makefile           <- Makefile with commands like `make data` or `make train`
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── docs               <- A default Sphinx project; see sphinx-doc.org for details
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
│                         the creator's initials, and a short `-` delimited description, e.g.
│                         `1.0-jqp-initial-data-exploration`.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
│
├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
├── ml4ir                <- Source code for use in this project.
│   ├── __init__.py    <- Makes ml4ir a Python module
│   │
│   ├── data           <- Scripts to download or generate data
│   │   └── make_dataset.py
│   │
│   ├── features       <- Scripts to turn raw data into features for modeling
│   │   └── build_features.py
│   │
│   ├── models         <- Scripts to train models and then use trained models to make
│   │   │                 predictions
│   │   ├── predict_model.py
│   │   └── train_model.py
│   │
│   └── visualization  <- Scripts to create exploratory and results oriented visualizations
│       └── visualize.py
│
└── tox.ini            <- tox file with settings for running tox; see tox.testrun.org

Project based on the cookiecutter data science project template. #cookiecutterdatascience