I would like to access the textual contents of another cell in the notebook from Python so that I can feed it to the unit testing script, regardless of whether it has been executed or not. It seems like this should be possible, but I can't find the right API to do it in the IPython kernel. Thoughts?
This question asks the same question, but I don't really want to use magics unless I have to. Ideally the workflow would be "select widget choice" followed by "click widget" to cause the tests to run.
Background
I'm working with some students to teach them Python, and in the past when I've done this I've set up a bunch of unit tests and then provided instructions on how to run the tests via a shell script. However, I'm working with some students that don't have access to computers at home, so I decided to try and use an Jupyter notebook environment (via mybinder.org) to allow them to do the same thing. I've got most of it working already via some ipywidgets and a helper script that runs the unit tests on some arbitrary set of code.
As far as the cells that have been run, the input and output caching system described at https://ipython.readthedocs.io/en/stable/interactive/reference.html#input-caching-system might be useful. (Examples of its use at https://stackoverflow.com/a/27952661/8508004 ). It works in Jupyter notebooks as shown below. (The corresponding notebook can be viewed/accessed here.)
Because #A. Donda raised the issue of markdown in the comments below, I'll add here that nbformat provides related abilities that works with saved notebook files. Reading a saved notebook file with nbformat allowa getting cells and content, no matter if it is code or markdown, and sorting whether the cells are markdown or code, etc.. I have posted a number of examples of using nbformat on the Jupyter Discourse forum that can be seen listed via this search here. It offers more utility than the related json.load(open('test.ipynb','r')) command, highlighted in a comment here to read a notebook file because of the additional notebook context automatically included.
If you want to capture the contents of a specific cell to access it from another one, one workaround seems to bee to write the contents of the first cell to file when executing it and later loading the resulting file (i.e. the text giving the former cell's content) inside the latter cell, where the content of the former cell is required.
Cell's contents can be saved to file (when executing the respective cell) via the command
%%writefile foo.py
which has to be placed at the beginning of a cell. This results in the cell's content (in which the upper command is executed) being saved to the file foo.py and it's just a matter of later reading it in, again.
The output of a cell can be made available more easily:
Just place %%capture output in the first line of a cell. Then, the output of the cell (after execution) is going to be saved as a string to the variable output and can be used like any standard python string-variable.
References:
Programmatically get current Ipython notebook cell output? and
https://nbviewer.jupyter.org/github/ipython/ipython/blob/1.x/examples/notebooks/Cell%20Magics.ipynb
I found a solution to this that can get the output of any cell. It requires running some Javascript.
Here is the code that you can put in a cell and run it an it will generate a Python variable called cell_outputs which will be an array of the cell outputs in the same order as they appear on the page. The output of the cell executing this code, will have an empty string.
%%js
{
let outputs=[...document.querySelectorAll(".cell")].map(
cell=> {
let output=cell.querySelector(".output_text")
if(output) return output.innerText
output=cell.querySelector(".rendered_html")
if(output) return output.innerHTML
return ""
}
)
IPython.notebook.kernel.execute("cell_outputs="+JSON.stringify(outputs))
}
If you need the output of a specific cell, just use its index, eg: cell_outputs[2] in your Python code to access the output of cell #3.
I tested this on my Jupyter notebook 6.0.3 (recent install via the Anaconda community edition) on Google Chrome. The above could should work fine on any modern browser (Chrome, Firefox or Edge).
Related
I want to know if there is an option in the Notebook UI or some magic command (for a cell) that prevents its re-execution.
For e.g., if I execute a few cells in sequence, and wanted to go back to re-run the first cell, then I should be prevented from doing that.
For the old Jupyter notebook see: How to freeze a cell alone with its outputs on a jupyter notebook.
For JupyterLab there is no extension yet but there is an agreement that this is a good idea. Everyone can become an extension author and contribute it. A good starting point is the Extension Developer guide.
If the cell has no output that you care about you can use a handy trick of temporarily switching its type to "raw" (as raw cells don't get executed). It can be done with a keyboard sequence of Esc, r; then to restore its type to "code" cell you can use Esc, y.
Python Markdown is a very nice extension for the jupyter notebook, which is in turn great for literate programming, i.e. mixing text and code.
Python Markdown makes it possible to include short inline code in Markdown cells in Jupyter like in the following example:
Python cell: a = 3.1415
Markdown cell: The value of a was {{a}}.
Everything works fine in the browser interface, but when I export it to LaTeX (or PDF via LaTeX), the output will still contain {{a}} as an unevaluated expression.
It would of course be really helpful to have the evaluated expression in the output for generating reports.
The solution was actually rather trivial:
When enabling a certain option on the command line, this will create an entry in the web interface irrespective of this option actually exists.
I had mispelled Python Markdown with python-markdown and ended up with this second entry in the web interface.
Enabling the first entry fixed the problem.
The second entry could be savely removed.
Is it possible to execute ONLY the highlighted code in a Jupyter notebook cell? This is possible in Spyder and RStudio.
I find this to be quite useful for trouble-shooting code as you write.
If a cell contains:
a=13
b=17
c=42
a=a*c
I'd like to be able to highlight and run only the desired lines (e.g. variable assignmemnts), but not the final line.
I use this frequently in Spyder and RStudio, would love to do in Jupyter as well. I find I am constantly splitting and re-combining cells in order to troubleshoot a single line of code, where for example, I've indexed into something incorrectly. Highlighting and printing the variable allows me to see what I've actually assigned it to be and is throwing an error, vs. what I had intended.
There is no such thing in Jupyter as 'highligh and run'. At least I am not aware of it.
Run the cell after commenting the other lines out using CTRL + /, split cells and execute only the chosen ones or use a debugger (e.g. pudb, it works in Jupyter) to change variables values on the fly (while debugging).
It seems now it is available in python notebook as well.
https://github.com/jupyterlab/jupyterlab/pull/2191
If I open a python notebook in Kaggle (www.kaggle.com) and select a text, it lets me run only the highlighted part.
I am currently creating many tables in python and most of the cells need their own SQL query. I am ending up with a long list of SQL querys. Ideally I would put the table creation of each table in its own file and load the files from a main notebook.
Is there a way to run sub-notebooks (containing the tables) from one main notebook? Mathematica has the EvaulateNotebook[] function which does exactly this. For python I found the nbconvert package which seems to come close to what I need. However, nbconvert seems to run in its own kernel. I would need the sub-notebook to have access to all variables in the main-notebook and subesquently the main-notebook to have acess to the variables in the sub-notebook.
In the end I would like to have the convenience of the notebooks in terms of editing, while not having to scroll through each and every table if I only need to change one table. So I am a bit reluctant to build a standard python package which I import, but would rather have a way to run a sub-ipynb in a way as if the cells where in the main-notebook.
I don't believe this is currently possible in Jupyter to have "child" notebooks in the same way you can with knitr.
Fernando Perez wrote a script to merge together multiple notebooks which may be helpful in your case to avoid having to scroll through a lot of text.
It looks like this may also eventually be incorporated into the main nbconvert codebase (see issue here).
I am trying to write an IPython notebook extension that will run analysis on the code in each cell and will then output messages via the logger (this could be for example a PEP8 analysis, to take a simple example). The catch is that I want to be able to load the extension once at the top of the notebook and have it apply to all cells.
It is possible to define custom line and cell magics (see here) but one has to put the magic command in each cell whereas I would like it to apply to all code cells.
I also found that it's possible to register a class that pre-filters code input, using ip.prefilter_manager.registre_transformer but that operates on a line by line basis which is not good enough.
Does anyone have any suggestions on how to register a function that can analyze the content of all cells, when they are executed, and add logger messages to the output of that given cell?
I know there is a PEP8 extension for IPython notebook but this requies that the magic %%pep8 command be put inside each cell to be checked, and I want to change this to be more of a global setting in the notebook.