Lesson 3 In-Class Discussion

[Chris_Palmer]

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

[pete.condon]

learn.predict(is_test=True)

[jenna]

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

[KevinB (Kevin Bird)]

Another treat! Early access to Intro To Machine Learning videos

We’ll be releasing a new course (tentatively) called Machine Learning For Coders towards the end of the year. It’s being recorded with the masters students at MSAN. I’ve decided to share the videos with you all as well, since for those of you that haven’t done any ML before, you might find this a helpful additional resource. It uses the same fastai library as our deep learning course, so your existing repo and AWS instance will work fine. Here’s the videos - I’ll update this thread as they’re a…

The files are in the ml1 folder in the github repo

[kmatsuda (Ken)]

@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.

[jenna]

Thank you

[KevinB (Kevin Bird)]

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

[yinterian]

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

[divyansh (Divyansh Jha)]

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

[kmatsuda (Ken)]

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.

[jenna]

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

[init_27 (Sanyam Bhutani)]

I don’t think so, It was unlisted.

[zaoyang]

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

[init_27 (Sanyam Bhutani)]

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

[yinterian]

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

[metachi (Jeff Lee)]

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%

[zpnc (Anze Zupanc)]

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

[guthl (Louis Guthmann)]

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

[zaoyang]

Intuitively every layer captures either edges, shapes of the image. As information passes through these kernels and activations, isn’t there progressive information loss? Is information preserved or does this not matter?

[tom1 (Tom Grek)]

Are there any situations where a non-square maxpool might be useful?