jdc - Jupyter Dynamic Classes¶

Alex Hagen

jdc is a Jupyter magic made to solve a very specific problem: that sometimes you want to define classes dynamically in Jupyter notebooks, and you want it to look good. On Jupyter’s Github there’s a lengthy discussion and wider view of this issue, but I’ve implemented a quickfix which I think ended up looking pretty semantic. Thanks to Igor Sobreira for the underlying IPython.

The example below will be most informative, but basically all jdc allows you to do is to add a cell and define

%%add_to our_class
def our_function(self, our_variable):
  print our_variable

and that will add the method our_function to our_class, whether our_class is a class, or an object with a class type.

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

Example¶

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

Below is an example on how to use jdc. First we have to import the class:

In [1]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

import jdc

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

Then, we have to make a dummy class:

In [2]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

class dog(object):
    def __init__(self, name, noise):
        self.name = name
        self.noise = noise


rover = dog('Rover', 'ruff')
spot = dog('Spot', 'woof')

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

Then, we want to add a function to that class:

In [3]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

%%add_to dog
def bark(self, times=1):
    saying = ('%s ' % self.noise) * times
    print("%s says: %s" % (self.name, saying))

def bark2(self, times=1):
    saying = ('%s ' % self.noise) * times
    print("%s says: %s" % (self.name, saying))

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

And now we can access that function from any method of that class.

In [4]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

rover.bark(1)
spot.bark(2)

Rover says: ruff
Spot says: woof woof

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

If we want to add a function to only one object of that class, we can do that, as well:

In [5]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

%%add_to spot
def sit(self):
    print('%s is now sitting' % self.name)

In [6]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

spot.sit()

Spot is now sitting

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

Note that the function added to a class object is only available to that object, not all objects of that class:

In [7]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

try:
    rover.sit()
except AttributeError:
    print("%s doesn't know that trick" % rover.name)

Rover doesn't know that trick

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

And if we’re writing a lot of code, we can now do that in a single cell.

In [8]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

%%add_to spot
def rollover(self):
    print("%s rolled over" % self.name)

def highfive(self):
    print("%s is trying to high five you" % self.name)

def domytaxes(self):
    print("%s is just showing off now" % self.name)

In [9]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

spot.rollover()
spot.highfive()
spot.domytaxes()

Spot rolled over
Spot is trying to high five you
Spot is just showing off now

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

Nesting functions and decorators should now work. Notice when calling the wrapped function that we must call it by func(self, *args) instead of self.func(*args) or func(*args).

In [10]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

%%add_to dog
def race(func):
    def wrapper(self):
        import time
        t1 = time.time()
        func(self)
        t2 = time.time()
        print("A new record time, %s finished in: " % self.name + str((t2 - t1)) + " seconds!\n")
    return wrapper

@race
def race_course(self):
    print("%s finished the course." % self.name)

In [11]:

jp-CodeMirrorEditor jp-Editor jp-InputArea-editor

spot.race_course()

Spot finished the course.
A new record time, Spot finished in: 5.984306335449219e-05 seconds!

jp-RenderedHTMLCommon jp-RenderedMarkdown jp-MarkdownOutput

and that’s it.

Installation¶

Right now, installation should now work through PyPI:

pip install jdc

Please open up an issue if it doesn’t work, I’ll fix it!

Note¶

@dsblank has pointed this out on the Jupyter discussion:

you can do this in regular Python, without resorting to any extra magic:

Cell 1:

class MyClass(): def method1(self): print(“method1”) Cell 2:

class MyClass(MyClass): def method2(self): print(“method2”) Cell 3:

instance = MyClass() instance.method1() instance.method2() That is, you can define a class recursively, cell by cell. I don’t think the final class is any different from one defined all in one cell. So you don’t need an extra package to “solve” this problem.

I personally like the semantics of jdc, so I won’t be removing it from PyPI, but it’s always good to know alternatives, thanks @dsblank!

Indices and tables¶

Contents: