Lesson 3 In-Class Discussion

Looks like he is answering the question.

1 Like

data.test_ds.fnames gives you the names of the files in the test in the same order as the predictions.

1 Like

I found TTA made my results slightly worse on the Dog Breeds experiment. Is there a way to get the probabilities without using TTA?

learn.predict(is_test=True)

2 Likes

Jeremy mentioned the fastai machine learning course? I’m having trouble finding it on the interwebs. Does someone have a link?

2 Likes

The files are in the ml1 folder in the github repo

3 Likes

@yinterian Thanks. Does the order of the files matter in the submission? In the dogs vs. cats one, I thought I was getting the files out of order.

Thank you :slight_smile:

They are awesome. I am only through lesson 2, but there is a ton of really good content.

1 Like

What matters is that you match the prediction with the right “id”.

why model is giving different predictions for single image?
Initial 37 and then 33?

Thank you. I think I remembered the order looking more straightforward in the version 1 keras version, but I think it was because it was pulling from the test directory sorted by the file system. I thought I had specified something incorrectly in the fastai library’s predict method.

Is Octavio’s video publicly available? (if so, link please <3)

5 Likes

I don’t think so, It was unlisted.

where’s the Octavio video? Is it available? Tried searching and couldn’t find it.

3 Likes

@yinterian
How do we arrive at the filters?
Are these optimised as well (via gradient descent or other methods?)

1 Like

Yes, the filters are optimized with Stochastic Gradient Descent (SGD) or a version of it.

2 Likes

I have a Keras implementation that looks a lot more like the fast ai lib. forum thread is here and github here.

Right now it gets about 98.5% acc on dogs and cats. I am still looking for the places where fastai (pytorch) and my lib differ to get it up to 99%

6 Likes

If @jeremy would have 3channel input in his excel spreadsheet, then the filter in first hidden layer would have dimension [3,3,3]?

Why using a simple sum of previous convolutions instead of weighted one?