tiwarilaxuu / pytorch-practice

Pytorch-Practice

What is pytorch?

-Both developer and researcher use

-Computer vision uses

-Replacement of numpy

-Very high performance on GPUs with CUDA

-Extensive neural network building blocks

-Built-in backpropagation with “autograd”

-A simple, intuitive API library

-Based on LUA programming language

-Simplicity so uses

Here, we used Softmax classifier gives a slightly more intuitive output (normalized class probabilities) and also has a probabilistic interpretation. Such networks are commonly trained under a log loss (or cross-entropy) regime, giving a non-linear variant of multinomial logistic regression. Here, we also classify MNIST classifier example. Mnist Dataset contains 0-9 character. It is 10 label problem and to predict one out of 10 classes we transform our above Y to have 10 discreet values i.e 10 outputs. From probability distribution, we just take the maximum probable one. It is magic, it select one class which has maximum probability and forgets others.