Which layer to choose - guideline

[OneManShow0815]

How do I choose a layer when for machine learning. Or, for what purpose is which layer best suitable?

[LuluW8071]

@OneManShow0815 Can you give more context

[OneManShow0815]

Hi LuluW0871, there a re many different Linear models in PyTorch. Every one has a different algorithm in it. Is there a description for which case files which Linear model suits the best?
Or after which guidlines do you choose a Layer when writing a software using PyTorch? Danial used ..nLinear in his first part, later on other ones.
As of now, I do not have specific task to write, just to understand how to choose one of the different layers

[LuluW8071]

There is no strict rule regarding the number of layers you should use in a model. Generally, linear models are utilized as feature selectors in the middle or final layers of CNNs and RNNs or in combinations of both. Additionally, sequential layers are used to stack linear or CNN layers for convenience, during the implementation of the forward pass function.

Since u r at the beginning of the course, understanding these concepts might be hard at first, but they will become clearer as u progress through the CNN (computer vision) section. I recommend reviewing the theory on artificial neural networks (ANNs) to grasp how linear networks function before diving into coding. Daniel has explained about ANNs as well before coding it.

[OneManShow0815]

Hi LuluW8071, Thanks for the quick response.
My question was more which layer y use, like Dropout layer or recurrent layer or other.

Regards
Sven

[LuluW8071]

Linear Layer for basic model or mid/final feature selector
Dropout layer for making our model unbiased towards training data only introducing non linearity
BatchNorm for stable (recenter and rescale with normlization) and fastertraining
CNN for image data(picture, audio)
RNNs for sequential(mostly textual and time-series) data
Transformer for text, images, videos, audios

Watch some vids and read research papers to get the concepts behind it

[OneManShow0815]

Thank you LuluW8071,

Great answer, 100% clear. Great.

Regards
Sven