Getting started with Jupyter - internals
Jupyter notebook is a web based application that uses web socket to talk to ipython, execute python (or other languages like Julia and Haskell). It is really popular when it comes to executing machine learning kernels.
Say for example, you trying to use numpy to read off certain csv files and runs mathplot, you can easily do so using IPython.
Great! So how does it work?
First you need to setup jupyter notebook. I assumed that you have installed Anaconda in your system.
First, lets setup our virtual environment
At the end of the cell execute request, results will be display / render on notebook. Communication is achieve via web socket and all the code is in javascript and not python. Will consume alot of time, if you start looking at the wrong place.
To see what happens when a user submit a python code say "print("testing testing "")"", code,
Take a look at notebook/js/services/kernel/kernel.js, in particular
1. execute function,
2. send_shell_message,
3. kernel._send - sends code over for execution.
Response flow (notebook/js/outputarea.js.
1. kernel._handle_output_message - receive response from ipython execution.
2. OutputArea.prototype.append_stream function - basically renders output to a div.
When you're exposed to the code above, you will be fine and able to understand the internals of jupyter.
'
You can examine it by looking at a javascript notebook/js/codecell.js under the method called "CodeCell.prototype.get_callbacks"
In fact, jupyter console uses the same approach.
Any responses from ipython -> jupiter client -> notebook gets display here.
What happens if you do a plot? The results will be return as embedded image like what you would normally do with based64 image encoder for your web page.
Say for example, you trying to use numpy to read off certain csv files and runs mathplot, you can easily do so using IPython.
Great! So how does it work?
First you need to setup jupyter notebook. I assumed that you have installed Anaconda in your system.
First, lets setup our virtual environment
Now activate our newly created environment.
pip install --upgrade setuptools pip
git clone https://github.com/jupyter/notebook
cd notebook
pip install -e .
From diagram below, we can see notebook depends on jupyter_client (purple) to talk to ipython (yellow)
At the end of the cell execute request, results will be display / render on notebook. Communication is achieve via web socket and all the code is in javascript and not python. Will consume alot of time, if you start looking at the wrong place.
To see what happens when a user submit a python code say "print("testing testing "")"", code,
Executing a piece of code.
1. execute function,
2. send_shell_message,
3. kernel._send - sends code over for execution.
Response flow (notebook/js/outputarea.js.
1. kernel._handle_output_message - receive response from ipython execution.
2. OutputArea.prototype.append_stream function - basically renders output to a div.
When you're exposed to the code above, you will be fine and able to understand the internals of jupyter.
'
Returning code execution results
User code execution results are obtain from IOPUB. Yeap, IOPUB, not SHELL socket.You can examine it by looking at a javascript notebook/js/codecell.js under the method called "CodeCell.prototype.get_callbacks"
In fact, jupyter console uses the same approach.
Any responses from ipython -> jupiter client -> notebook gets display here.
What happens if you do a plot? The results will be return as embedded image like what you would normally do with based64 image encoder for your web page.
Comments