pytorch in 1 hour

.shape → a tuple describing the dimensions #1 debugging tool .device → where the tensor lives. cpu or cuda (GPU) .dtype → the data type of the numbers. The default is float32 (because of backprop and gradients)

model weights and biases must be a float32 (standard) data that represents categories or counts can be integers

Autograd - Automatic Differentiation

it is python’s built in gradient calculator, and can be turned on using requires_grad=True
to tell pytorch a tensor is a learnable param we must set requires_grad=True, doing so pytorch tracks every single operation on that tensor!

The difference between * and @ in pytorch

for @ multiplication m1 col = m2 rows - used for neural networks
when building a linear layer always use @ → y = X@W + b

Reduction - any operation that reduces a tensor to a smaller number of elements ex- sum(), mean(), max() etc.

dim arg

selecting data - basic and custom

making forward pass

model’s first guess
simple linear regression ŷ = XW + b The magic command - loss.backward()
- calculates the gradient of the loss wrt to our weight “W”
- the gradient of the loss wrt to our Bias ‘b’ for all the terms that are present in the autograd graph that pytorch has built!