Good resources re: deployment in general (for beginners)?
I have done both the courses. But I felt it to be more theoretical and I was not confident enough to dive into dive into Deep Learning coding as opposed to this course. But I think it was also a great course to get deeper knowledge
I think doing a simple Flask tutorial would be helpful for understanding deployment: http://flask.pocoo.org/docs/1.0/quickstart/
Once you see how to set up a simple webserver that replies with “hello world”, you’ll just need to adapt for response that returns your model prediction. Here’s my example
Anyone have resources for how to get started using video? (i.e. pulling frames from a video and submitting those to a model?)
I made a post earlier this year before V1 came out and didn’t get much traction. I would love to train a model and build an app to play the game Set. It’s a fantastic card game and even Peter Norvig did an analysis of it.
Anybody want to form a team? I think the hardest part of this will be the dataset. I’m not in a study group at this point as I’m remote and kind of alone in my city.
Where to find today’s class slides?
Can we set how many augmentation we apply for one image in epoch?
I write a short tutorial to use kaggle API in google colab. It may help
Getting a bunch of errors. E.g., in Config.data_path()/'planet'
, it’s saying that Config
doesn’t have data_path
attribute. Similarly
src = (ImageFileList.from_folder(path)
.label_from_csv('train_v2.csv', sep=' ', folder='train-jpg', suffix='.jpg')
.random_split_by_pct(0.2))
fails with
TypeError: datasets() missing 1 required positional argument: 'dataset_cls'
When your model makes an incorrect prediction in a deployed app, is there a good way to “record” that error and use that learning to improve the model in a more targeted way?
Is there a way to predict a class and a number?
First you need to make a model that outputs two things - a vector of class predictions and the number. Then you need a loss function that evaluates both.
Those are the sames as the transforms applied on pets, with just vertical flips and rotations of 90 degrees on top of the horizontal flip.
Does it accept mulitple extensions, like .jpeg, .png. gif ?
You filenames need to end with them in that case. Here the .jpg was missing in all the filenames.
Anyone else get Type Error: datasets() missing 1 required positional argument: ‘Dataset_cls’ when this is called:
data=(src.datasets().transform...)
What if some of my images are PNG and some are JPEG and so on…
My guess is that you don’t have the latest version of the library.
Like I said, the csv approach will work as long as you have the full filenames there.