Trying to import movie_recommender example function onto index 'nn' is not defined

nbdev
1

I’ve created and posted the Fastai movie recommender using nbdev HERE where the recommender worked nicely.

I’m trying to add the get_movie_recs function from the 00_movie_recommender.ipynb to the index.ipynb file so it shows up on the README.

Despite using:

from movie_recommender.movie_recommender import *
from fastai.tabular.all import *
from fastai.collab import *

Running get_movie_recs function on index.ipynb throws:

"NameError: name 'nn' is not defined"

I’ve tried adding torch to the ‘requirements’ in settings.ini to no avail. How can I get index.ipynb to recognize the ‘nn’ from nn.CosineSimilarity?

Any help would be greatly appreciated.

Screen Shot 2022-10-14 at 10.27.31 AM
Screen Shot 2022-10-14 at 10.27.31 AM2044×760 133 KB

2

Very strange, how about

import torch.nn as nn
3

Thanks for the idea, but it didn’t work. I added :

import torch.nn as nn

to both .ipynb files and it still will not run on the index. I can copy and paste the whole function into the index notebook, but that seems to defeat the purpose right?

4

Yeah whole function isn’t ideal.

Is pytorch installed?

If not could pip install it

!pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html
5

so fastai apparently already needs torch to be loaded which is why nn.Cosin… works in the movie_recommender.ipynb file. I tried your pip install but this(below) happpend and it still throws the error: 'nn' is not defined

Screen Shot 2022-10-14 at 2.56.11 PM
Screen Shot 2022-10-14 at 2.56.11 PM1294×1216 204 KB

6

Could you try importing torch & torch.nn as below?

import torch
import torch.nn as nn
7

@daveramseymusic Hey, you are not exporting your imports. In particular, when you check movie-recommender/movie-recommender.py you can see that it doesnt import anything, so it can not know of nn. Add #| export to the top of

# For modeling
from fastai.tabular.all import *
from fastai.collab import *
#for publishing the model
import gradio as gr

and it should solve your problem :slight_smile:

Edit: Great project btw, it picked my movie for tonight :laughing:

2 Likes
8

Aha! That worked nicely.

9

Glad to hear about the movie too. I’m about to load a larger 20mil reviews version in an hour.

10

Is it possible to import the learn object with the get_movie_recs() function?

In order to get the function to run simply on in the gradio block I must add learn=learn in the function parameters.

Now when I run: from movie_recommender.movie_recommender import *

This is the error:

NameError: name 'learn' is not defined

I’ve even saved the object into the index.ipynb notebook before the import to no avail using :

learn = load_learner(path/'movie_predictor_large20mil_639.pkl')

Is there a better way to approach this? Ideally, I wouldn’t have redundant code in the index.ipynb just to show the function in the README.

11

Update: I fixed the error: “name ‘learn’ is not defined” by adding #| export to the learn=load_learner(path/'movie_predictor_large20mil_639.pkl') cell.

12

In case this is useful for anyone else:

I was having difficulty running nbdev_prepare because the README was not in the nbs directory.

I fixed it using this code where I loaded the learn and titles:

#| export
#load the model and movie titles with indexes (from the previous data loaders)
path = Path('.')
data_folder = '../models_and_dls'

#check if modedels folder is up one level if not 'for readme' then check in current directory
if not os.path.isdir(data_folder):
    data_folder = './models_and_dls'
#load learner and titles csv    
learn = load_learner(path/data_folder/'movie_predictor_large20mil_xtrau_624.pkl')
titles = pd.read_csv(path/data_folder/'movie_20mil_xtrau_dls.csv')
1 Like
13

I wrote here about how to work with repo-relative paths, you might find it helpful:

2 Likes