From DeepSense Docs
Revision as of 11:15, 18 December 2020 by Lyang (talk | contribs) (Using Dash on DeepSense)
Jump to: navigation, search

In this section, we would not introduce those popular plotting libraries like Matplotlib, Pandas, Seaborn, ggplot, Plotly, and so on. We would try to introduce and provide information about some visualization frameworks and libraries that have more features. Using the features in these frameworks and libraries, you may make your data more understandable and interactive.



We would make use of the introduction of Dash directly from Dash's official webpage. Users can obtain more details about Dash accessing the official webpage.
"Dash is a productive Python framework for building web analytic applications.
Written on top of Flask, Plotly.js, and React.js, Dash is ideal for building data visualization apps with highly custom user interfaces in pure Python. It's particularly suited for anyone who works with data in Python.
Through a couple of simple patterns, Dash abstracts away all of the technologies and protocols that are required to build an interactive web-based application. Dash is simple enough that you can bind a user interface around your Python code in an afternoon.
Dash apps are rendered in the web browser. You can deploy your apps to servers and then share them through URLs. Since Dash apps are viewed in the web browser, Dash is inherently cross-platform and mobile ready.
There is a lot behind the framework. To learn more about how it is built and what motivated Dash, watch our talk from Plotcon below or read our announcement letter.
Dash is an open source library, released under the permissive MIT license. Plotly develops Dash and offers a platform for managing Dash apps in an enterprise environment.

Users can easily install Dash in their home directory on DeepSense. The way to install it is very similar to installing a package using conda. Here we provide how to install and use Dash on DeepSense. The install process only takes about 10 minutes if you have already set up your conda environments in your home directory on DeepSense.

Installing Dash

Activate the conda environment you would install Dash

[luy@ds-lg-01 ~]$ conda activate py36-pytorch

Running the following command to install Dash:

(py36-pytorch) [luy@ds-lg-01 ~]$ conda install -c conda-forge dash

After the installation is done, issue the following command to check if Dash is installed:

(py36-pytorch) [luy@ds-lg-01 ~]$ conda list |grep dash
dash                      1.18.1             pyhd8ed1ab_0    conda-forge
dash-core-components      1.14.1             pyhd8ed1ab_0    conda-forge
dash-html-components      1.1.1              pyh9f0ad1d_0    conda-forge
dash-renderer             1.8.3              pyhd8ed1ab_0    conda-forge
dash-table                4.11.1             pyhd8ed1ab_0    conda-forge

Learning Dash

Users can learn Dash on its official webpage. There are also a lot of Dash tutorials on youtube. You would be able to see how powerful Dash can do for visualizing your data.

Using Dash on DeepSense

1. Creating a sample Python script using Dash
Now you can develop your projects using Dash. Below is a simple example script showing how Dash works on DeepSense platform. Users can copy and paste the contents of the script and create a Python script in your home directory on DeepSense. Let's name it ''.

import torch
import torchvision
from platform import python_version
import dash
import dash_html_components as html
import subprocess
app = dash.Dash(__name__)
gpu_detail=str(['nvidia-smi'], stdout=subprocess.PIPE).stdout.decode('utf-8'))
gpu_name=' '.join(gpu_detail.split('\n')[7].split()[2:4])
   html.H6(children='Hello world!'),
   html.H6(children="Python Version: " + str(python_version())),
   html.H6(children="PyTorch Version: " + str(torch.__version__)),
   html.H6(children="Torchvision Version: " + str(torchvision.__version__)),
   html.H6(children="GPU name: " + gpu_name)]
app.layout = html.Div(children=content)
if __name__ == '__main__':

2. Getting an interactive session via LSF
Running the following command to submit a new LSF interactive job:

(py36-pytorch) [luy@ds-lg-01 ~]$ bsub -Is -gpu - bash
Job <6008> is submitted to queue <gpu>.
<<Waiting for dispatch ...>>
<<Starting on ds-cmgpu-01>>
[luy@ds-cmgpu-01 ~]$

Now you have an interactive session opened on the compute host ds-cmgpu-01. You can run the sample script on the command line:

[luy@ds-cmgpu-01 ~]$ python 
Dash is running on
* Serving Flask app "dashTest" (lazy loading)
* Environment: production
  WARNING: This is a development server. Do not use it in a production deployment.
  Use a production WSGI server instead.
* Debug mode: off
* Running on (Press CTRL+C to quit)

3. See the output on a webpage
Keep the session open. This means you should not press CTRL+C. You would need to do port forwarding to open the webpage on the web browser on your local machine. This is very similar to how to use notebook on DeepSense. Below is an example:

Lus-MacBook-Pro:~ lyang$ ssh -l luy -L 8050:ds-cmgpu-01:8050's password: 
Last login: Fri Dec 18 08:41:15 2020 from
IMPORTANT: Dear DeepSense users, jobs running on the head nodes (login 
node 1 and login node 2) must be less than 10 minutes and have small 
memory and CPU requirements. Jobs violating this will be terminated 
without warning.
[luy@ds-lg-01 ~]$ 

Be careful that we obtain the compute host name 'ds-cmgpu-01' and port number '8050' from step 2. The host name and port number may change when you submit your jobs. After the port forwarding is finished, you can copy the url '' found at the last line of step 2 and paste it to your local web browser. The above example script would output the following contents on a HTML web page:<br.

Hello world!
Python Version: 3.6.12
PyTorch Version: 1.6.0a0+445c276
Torchvision Version: 0.7.0a0
GPU name: Tesla P100-SXM2...

Please feel free to contact DeepSense at for any questions.