Coresets On Bacalhau

Introduction

Coreset is a data subsetting method. Since the uncompressed datasets can get very large when compressed, it becomes much harder to train them as training time increases with the dataset size. To reduce training time and cut costs, we employ the coreset method; the coreset method can also be applied to other datasets. In this case, we use the coreset method which can lead to a fast speed in solving the k-means problem among the big data with high accuracy in the meantime.

We construct a small coreset for arbitrary shapes of numerical data with a decent time cost. The implementation was mainly based on the coreset construction algorithm that was proposed by Braverman et al. (SODA 2021).

For a deeper understanding of the core concepts, it's recommended to explore: 1. Coresets for Ordered Weighted Clustering 2. Efficient Implementation of Coreset-based K-Means Methods

In this tutorial example, we will run compressed dataset with Bacalhau

Prerequisite​

To get started, you need to install the Bacalhau client, see more information here

Running Locally​

Clone the repo which contains the code

git clone https://github.com/js-ts/Coreset

Downloading the dataset​

To download the dataset you should open Street Map, which is a public repository that aims to generate and distribute accessible geographic data for the whole world. Basically, it supplies detailed position information, including the longitude and latitude of the places around the world.

The dataset is a osm.pbf (compressed format for .osm file), the file can be downloaded from Geofabrik Download Server

wget https://download.geofabrik.de/europe/monaco-latest.osm.pbf

Installing Dependencies​

The following command is installing Linux dependencies:

sudo apt-get -y update \
sudo apt-get -y install osmium-tool \
sudo apt-get -y install libpq-dev gdal-bin libgdal-dev libxml2-dev libxslt-dev

Ensure that the requirements.txt file contains the following dependencies:

# requirements.txt
certifi==2020.12.5
chardet==4.0.0
cycler==0.10.0
idna==2.10
joblib
kiwisolver==1.3.1
lxml==4.6.2
matplotlib==3.3.3
numpy==1.19.4
overpy==0.4
pandas==1.1.4
Pillow==8.0.1
pyparsing==2.4.7
python-dateutil==2.8.1
pytz==2020.4
requests==2.25.1
scikit-learn
scipy
six==1.15.0
threadpoolctl
tqdm==4.56.0
urllib3==1.26.2
geopandas

The following command is installing Python dependencies:

pip3 install -r Coreset/requirements.txt

Running the Script​

To run coreset locally, you need to convert from compressed pbf format to geojson format:

osmium export monaco-latest.osm.pbf -o monaco-latest.geojson

The following command is to run the Python script to generate the coreset:

python Coreset/python/coreset.py -f monaco-latest.geojson

coreset.py contains the following script here

Containerize Script using Docker​

To build your own docker container, create a Dockerfile, which contains instructions on how the image will be built, and what extra requirements will be included.

FROM python:3.8

RUN apt-get -y update && apt-get -y install osmium-tool && apt update && apt-get -y install libpq-dev gdal-bin libgdal-dev libxml2-dev libxslt-dev

ADD Coreset Coreset

ADD monaco-latest.geojson .

RUN cd Coreset && pip3 install -r requirements.txt

We will use the python:3.8 image, we run the same commands for installing dependencies that we used locally.

See more information on how to containerize your script/app here

Build the container​

We will run docker build command to build the container:

docker build -t <hub-user>/<repo-name>:<tag> .

Before running the command replace:

hub-user with your docker hub username, If you don’t have a docker hub account follow these instructions to create docker account, and use the username of the account you created

repo-name with the name of the container, you can name it anything you want

tag this is not required but you can use the latest tag

In our case:

docker build -t jsace/coreset

Push the container​

Next, upload the image to the registry. This can be done by using the Docker hub username, repo name or tag.

docker push <hub-user>/<repo-name>:<tag>

In our case:

docker push jsace/coreset

Running a Bacalhau Job​

After the repo image has been pushed to Docker Hub, we can now use the container for running on Bacalhau. We've already converted the monaco-latest.osm.pbf file from compressed pbf format to geojson format here. To submit a job, run the following Bacalhau command:

bacalhau docker run \
    --input https://github.com/js-ts/Coreset/blob/master/monaco-latest.geojson \
    jsace/coreset \
    -- /bin/bash -c 'python Coreset/python/coreset.py -f monaco-latest.geojson -o outputs'

Structure of the command​

Let's look closely at the command above:

  1. bacalhau docker run: call to bacalhau

  2. --input https://github.com/js-ts/Coreset/blob/master/monaco-latest.geojson: mount the monaco-latest.geojson file inside the container so it can be used by the script

  3. jsace/coreset: the name of the docker image we are using

  4. python Coreset/python/coreset.py -f monaco-latest.geojson -o outputs: the script initializes cluster centers, creates a coreset using these centers, and saves the results to the specified folder.

Additional parameters:

-k: amount of initialized centers (default=5)

-n: size of coreset (default=50)

-o: the output folder

When a job is submitted, Bacalhau prints out the related job_id. We store that in an environment variable so that we can reuse it later on.

Declarative job description​

The same job can be presented in the declarative format. In this case, the description will look like this:

name: Coresets On Bacalhau
type: batch
count: 1
tasks:
  - name: My main task
    Engine:
      type: docker
      params:
        Image: "jsace/coreset" 
        Entrypoint:
          - /bin/bash
        Parameters:
          - -c
          - "osmium export input/liechtenstein-latest.osm.pbf -o /liechtenstein-latest.geojson;python Coreset/python/coreset.py -f /liechtenstein-latest.geojson -o /outputs"
    Publisher:
      Type: ipfs
    ResultPaths:
      - Name: outputs
        Path: /outputs      
    InputSources:
      - Source:
          Type: "s3"
          Params:
            Bucket: "coreset"
            Key: "*"
            Region: "us-east-1"
        Target: "/input"    

The job description should be saved in .yaml format, e.g. coreset.yaml, and then run with the command:

bacalhau job run coreset.yaml

Checking the State of your Jobs​

Job status: You can check the status of the job using bacalhau job list.

bacalhau job list --id-filter ${JOB_ID} --wide

When it says Published or Completed, that means the job is done, and we can get the results.

Job information: You can find out more information about your job by using bacalhau job describe.

bacalhau job describe ${JOB_ID}

Job download: You can download your job results directly by using bacalhau job get. Alternatively, you can choose to create a directory to store your results. In the command below, we created a directory (results) and downloaded our job output to be stored in that directory.

rm -rf results && mkdir -p results
bacalhau job get $JOB_ID --output-dir results

Viewing your Job Output​

To view the file, run the following command:

ls results/outputs

centers.csv                       coreset-weights-monaco-latest.csv
coreset-values-monaco-latest.csv  ids.csv

To view the output as a CSV file, run:

cat results/outputs/centers.csv | head -n 10

lat,long
7.423843975787508,43.730621154072196
7.4252607,43.7399135
7.411026970571964,43.72937671121925
7.459404485446199,43.62065587026715
7.429551373022234,43.74042043301333
cat results/outputs/coreset-values-monaco-latest.csv | head -n 10

7.418849799999999384e+00,4.372759140000000144e+01
7.416779063194204547e+00,4.373053835217195484e+01
7.422073648233502574e+00,4.374059957604499971e+01
7.434173206469590234e+00,4.374591689556921636e+01
7.417540100000000081e+00,4.372501400000000160e+01
7.427359010538406636e+00,4.374324133692341121e+01
7.427839200000001085e+00,4.374025220000000758e+01
7.418834173612560257e+00,4.372760402368248833e+01
7.416381731248183229e+00,4.373708812663696932e+01
7.412050699999999992e+00,4.372842109999999849e+01
cat results/outputs/coreset-weights-monaco-latest.csv | head -n 10

7.704359156916230233e+01
2.090893934427382987e+02
1.560611140982714744e+02
2.516557569411126281e+02
7.714605094768158722e+01
2.640808776415075840e+02
2.326085291610944523e+02
7.704841021255269595e+01
2.089705263763523249e+02
1.728105655128551632e+02

Support​

If you have questions or need support or guidance, please reach out to the Bacalhau team via Slack (#general channel).

Last updated