I found that if I restart the kernel so I can train much faster. I think the reason is because of the GPU memory. GPU memory not being freed after training is over . I will read to see how to free the memory without restarting the kernel
1 Like
What human Human overfitting feels like 
1 Like
After banging my head around this for a lot of time, couldn’t identify the issue causing this unusual behaviour. Did a system reboot and looks like that has fixed it
Yeah I also have problems understanding it. I referred to the dev_nb in the fastai_old repo. But couldn’t get myself to understand it. It will help a lot if Jeremy tells about them in the coming lessons
1 Like
I noticed that Lr_find() in version 1.0 is significantly slower than the previous one. I also noticed that it runs 4 times, so does that mean it reduces lr for each mini batch through 4 epochs? I’m confused
The lesson 3 head-pose notebook currently gives out an error when the data object is created
@jm0077 has created an error trace
The notebook worked fine until I pulled and conda installed the new updates (fastai version 1.0.27)
I pondered about YouTube’s auto-caption feature while watching lesson-4 (NLP).
As per my guess, I think they use speech to text ML models to generate those captions.
Is there any possibility that in-addition they might also be using NLP to enhance the prediction ?
I think speech-recognition + NPL would really further improve the auto-caption performance.
Just want to go on record saying I LOVE Jeremy’s Excel demos. Not that I’m a major Excel-worker, but I find it so clearly emphasizes “what’s really going on” without any possible “magic”.
6 Likes
Do bias are also updated by product of learning rate with it??
b = b - db * learning_rate
or just this
b = b- db
The first one, like all parameters that are trainable.
1 Like
This might be a relatively basic question, but why is the batch size parameter always set to values that are powers of 2?
How to use test csv file with learn.predict() ?
How do we fetch the list of filenames from our dataset with the new updates to the library?
I’m running on planet-amazon. 10 days ago I was able to just call:
idx2class = {v:k for k,v in learn.data.train_ds.ds.class2idx.items()}
to convert class indices back to class-names, and:
fnames = [f.name.split('.')[0] for f in learn.data.test_ds.ds.x]
to get the filenames. But the .data.<xyz>_ds no longer has the .ds attribute I was using.
I checked the changelog and searched around the forums, but anything I found was from about a month ago. I’ll edit this post with the answer if I find it.
edit:
so looks like you can call:
learn.data.train_ds.x.items
to get the list of filepaths (and also for .valid_ds and .test_ds).
Is this the ‘right’ way to do it? And is this guaranteed to match up with predictions on the validation and test sets?
edit2:
Think I found how to get your class-to-index mapping:
learn.data.train_ds.y.c2i
It’s gotten more intuitive: (“where can I find filenames?” → take a look at where the data comes from: .<blah>_ds.x.<blah>; “where can I find how classes are one-hot encoded?” → look where the labels are stored: <blah>_ds.y.<blah>
2 Likes
Is there anyway I can read zip file directly through fast.ai libs and read all image files within zip files?
Thanks
Amit
I think there’s no fastai method to extract zip files right now, you have to decompress it on your own using an other library first
1 Like
Thanks…Have now used python zip module to unzip
unfreeze() function generally unfreezes all layer. I could not locate any fast.ai function which allows me to unfreeze last x layers and not all the layers…Can anyone please help me on this?
Thanks
Amit
look at the imdb notebook. I think you want to use
learn.freeze_to(-x) where X would be you last few layers or your choosing
3 Likes
A note: It looks like freeze_to freezes by layer group (learn.layer_groups). For resnet34 and resnet50 there are 3 layer groups.
So for finer control I guess you can just take a look at the function and apply
if not self.train_bn or not isinstance(l, bn_types): requires_grad(l, False)
on the layers (l's) you want to freeze.
I always insert in my notebooks fastai.__version__
1 Like