Kwargs in documentation

fastai fastai dev
1

the documentation references kwargs -> any, but i can’t find the documentation where kwargs are listed and which apply to which function (if it varies). Can someone point me to the location in the documentation? Thanks a ton

1 Like
2

kwargs is a python concept and I wouldn’t expect fast.ai doc to describe it

3

Yes kwargs are a python concept. fast.ai uses the concept in its functions to allow keyword arguments to be entered into the method call.
However, it is not clear which keywords which function accepts.
cnn_learner accepts the kwarg “model_dir” from the learner class.
The function predict also accepts kwargs, but probably not the same as cnn_learner. predict has no use or concept of a model_dir, which originates in the learner object that cnn_learner creates.

So my question still stands, where can i find the corresponding kwargs allowed in a specific function call? For example predict from the vision module.
Or is there a python generic way to identify these kwargs? Then that would obviously also answer the question. Your answer sadly does not :frowning:

4

Did you check docs.fast.ai? There you can find examples and links to test scripts as further examples

5

Yes i have checked docs.fast.ai (I literally started this post with “the documentation[…]”.)

cnn_learner has no tests.
And the source code from the predict test case does not show me the kwargs i can use with it.

def test_models_meta(mnist_tiny, arch, zero_image):
    learn = cnn_learner(mnist_tiny, arch, metrics=[accuracy, error_rate])
    this_tests(learn.predict)
    pred = learn.predict(zero_image)
    assert pred is not None
6

I have the same question. How do you make out which function accepts which all kwargs?
Any explanation? I went throu docs but kwargs not explained there for all functions.

For Eg, data.show_batch(). It accepts figsize as a keyword argument. But I am not able to find where is it documented.

7

some methods, like ImageDataBunch.from_folder actually have a reference. This is what i expected for all methods.

Refer to create_from_ll to see all the **kwargs arguments.

8

kwargs get passed down to the functions used by the function you’re calling. the data.show_batch(), for example, will pass the kwargs to self.train_ds.x.show_xys() and its show_xys() that can make use of figsize. Sometimes the documentation mentions where the kwargs will go. The cnn_learner() documentation says, for example, that kwargs are passed down to Learner. I found it easier (and quite helpful anyway) to read the code to see where kwargs go. Always learn something else along the way :slight_smile:

2 Likes
9

I agree with OP - the documentation could be much clearer in letting us know what the valid kwarg options are. I’m finding this quite frustrating - appreciate some more helpful tips if anyone has any other tips