Looking for documentation on 'path' argument in TabularDataBunch.from_df()

fastai
1

I’m not sure I understand the purpose of path (mandatory) argument in TabularDataBunch.from_df(path=path,df=df,...) of fast.ai library in Python 3.6.

I checked documentation, but can’t seem to find the details there.
In particular, I have a pd.DataFrame that does not have an associated CSV file on a disk. How do I go about applying .from_df method to it?

Does anyone have more info or links to references?

2

Found an example here that helped, but still would like to figure out what path argument is and how to use it.

df = pd.DataFrame({'A': list('aabbccabca'), 'B': np.random.normal(size=10).round(2), 'Y': list('aabbccabca')})
tfms = [Categorify]
tblrData = TabularDataBunch.from_df('output', df, dep_var='Y', valid_idx=[7,8], procs=tfms, cat_names=['A'], bs=4)
(cat_x,cont_x),y = next(iter(tblrData.train_dl))
for o in (cat_x, cont_x, y): print(to_np(o[:5]))

Also, is there a documentation about the bs argument?
UPDATE: found it in the video. bs is a batch size parameter.

3

path is just the working directory where temporary files/models will be saved.

4

Thanks Sylvain. I also found it in the video (lecture 4, 43rd minute).

Is there a documentation that describes this? I’m sure I will face more questions about other arguments as I start applying the methods.

5

Be careful, you’re using fastai v1 with the old MOOC so it won’t work properly.

6

My version of fastai is 1.0.39. What’s the best and up to date source explaining the usage and examples of the library?

7

Apart from docs.fast.ai and examples/notebooks on the fastai github (and the code itself), this is a good read on the datablock api: https://medium.com/@wgilliam/finding-data-block-nirvana-a-journey-through-the-fastai-data-block-api-c38210537fe4
and this tutorial on torch.nn by @jeremy will help put things in context https://pytorch.org/tutorials/beginner/nn_tutorial.html

1 Like