Lesson 4 In-Class Discussion ✅

Can we use VS Code on a remote machines?

If any of you have a good guide to use the features Jeremy just alluded to in vs code that’d be great !

Does anybody from fastai team use a visual debugger or IDE? It seems like you’re all perfectly fine with Vim and Jupyter =) Do you prefer set_trace (or breakpoint) calls instead of using UI tools?

4 Likes

squeeze is a PyTorch method to remove a dimension.

4 Likes

Any intuition about what this neural network learns in each intermediate layer (like the visualization he showed for CNN)? - that would help us understand what representation the network is building/improving…

3 Likes

same question for NLP?

@rachel I hope didn’t forget this one! :slight_smile:

1 Like

Not unless you’re Remote-Desktop-ing into it where you can see the graphical user interface. VSCode is a windowed, GUI app; not a terminal app.

Would X11 help?

Thanks for the reminder, I did forget

1 Like
13 Likes

Wow this is super useful!

1 Like

@jeremy did a text classifier on IMDB.
My question is how can we do a text regressor?

1 Like

I am not too sure we will cover character-level NLP model in this version of the course. In part1-v2, we learnt about this topic in lesson 6—Recurrent Neural Network(RNN), “Character based model vs. word based model”

but this time, we will do it one character at a time. RNN is no different from what we have already learned in lesson 4.

However, in part1-v2, we have not started using transfer learning technique for NLP approach. Note, there is also something in between word & char-level model that is called Byte Pair Encoding (BPE) which looks at n-gram of characters.

I’ll wait for the next lessons.

1 Like

Why no activation function for the last output?

What do you mean with ‘text regressor’?

@rachel Requesting again! :slight_smile:

1 Like

model input is text, model output is a number.

1 Like

You mean continuous number? Can you give an example?

Do you mean activation function ?