1 - "pytorch deep learning practice" - linear model

Reading a book , With your heart , Summary methodology , application
data , Model , Training , application

Linear model

Training

Training test

Training test , Limited samples may not be able to fit the true distribution of all data

Over fitting , Recite the answer

It is divided into a development set or a verification set

Like the simplest way , Try it first

W Or in a specific way , That's the most suitable , You can try one first , Take action first

Only when you take action can you make adjustments , Continuous iterative optimization and adjustment according to the error image

Loss function , What's the difference , Refers to the distance , Use square or absolute value , Square differentiable

Find the minimum

Loss function , For a sample , Evaluate the loss function for the whole sample ,MSE


The curve is the result of application and persuasion

Random range to test , Visually find the minimum value

This is the exhaustive method

Visualization of training losses , Log or visualization tool

visdom,facebook Visual Open Source Toolkit , Direct landing web service , Watch the training in real time , Including breakpoint visualization

Multi dimensional parameters , Surfaces