Skip to content
AnacondaCon 2019 Dask for ML talk using ArXiv NLP Case Study
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
assets
ArXiv NLP Case Study.ipynb
Dockerfile
README.md
docker-compose.yml
start.sh

README.md

Anacondacon-2019-talk

Dask for ML Workflows code examples

1. Clone this repo

2. Launch docker container

2a. Pre-requisites

Docker installed on your machine

If you are working on mac, it is recommended to increase the resources the docker-machine gets from the host operating system. The default resource limits can cause memory issues and the ipython kernel may restart during Dask operations. This is easily done through the Docker App GUI interface. See the screenshot below

2b. Setup

./start.sh

docker ps -> should launch an image like below

docker ps

Note down the PORTS information. We map the default jupyter notebook port 8888 and Dask's dashboard port 8787 to ports on your host machine that can be accessed.

3. Notebook

3a. Pre-requisites

PORT from docker ps output

3b. Open notebook server

In your browser go to the following address:

localhost:<PORT>

where PORT corresponds to the following value in PORTS:

127.0.0.1:<PORT>->8888/tcp, a value near 32768

In the example above, this would be 127.0.0.1:32768->8888/tcp, so we would go to localhost:32768

3c. Open notebook

ArXiv NLP Case Study.ipynb can be found at the top level in this repo

3d. Connect to Dask dashboard

Dask provides a nice dashboard to monitor the cluster. This is usually available on port 8787. We mapped the container's 8787 to a port on the host. The cell with following lines initializes a Dask cluster.

` cluster = LocalCluster()

client = Client(cluster)

client `

After your ran that notebook cell you can connect to the dashboard page by going to

localhost:<PORT>/status

where PORT corresponds to the following value in PORTS:

127.0.0.1:<PORT>->8787/tcp, a value near 32768

In the example above, this would be 127.0.0.1:32769->8787/tcp, so we would go to localhost:32769/status

When you are executing a dask operation, the dashboard may look like below which provides information on running tasks and memory usage:

dask dashboard

You can’t perform that action at this time.