In the latest master branch, ConvLearner has been replaced with create_cnn
ConvLearner(data, models.resnet34, metrics=error_rate)
now becomes
create_cnn(data, models.resnet34, metrics=error_rate)
Yup - everyone should ensure that they notifications turned on for the official updates thread, since any breaking changes are announced there:
2 Likes
It works now after I issued the following commands also:
conda install -c pytorch pytorch-nightly cuda92
conda install -c fastai torchvision-nightly
conda install -c fastai fastai
1 Like
When returning to work, follow all the steps at http://course-v3.fast.ai :
1 Like
I’m using Developer Install and to fix it i’ve run first:
conda install -c pytorch pytorch-nightly cuda92
conda install -c fastai torchvision-nightly
and then inside fastai root folder:
pip install -e .[dev]
Otherwise I’ve got this error:
...
~/work/fastai/fastai/basic_train.py in on_train_begin(self, pbar, metrics_names, **kwargs)
241 self.pbar = pbar
242 self.names = ['epoch', 'train_loss', 'valid_loss'] + metrics_names
--> 243 self.pbar.write(' '.join(self.names), table=True)
244 self.losses,self.val_losses,self.lrs,self.moms,self.metrics,self.nb_batches = [],[],[],[],[],[]
245
TypeError: write() got an unexpected keyword argument 'table'
As suggested by @smugger, updating fastprogress will fix the issue.
With a dev install you have to manually update deps. In this case, your fastprogress lib is out of date.
Is there a way to just update the packages associated to a specific channel?
Tried: conda update -c fastai --override-channels --all … but if failed to pick up the lastest fastprogress package. I had to update that one individually via conda update -c fastai fastprogress.
Thanks