# Write an ANN from Scratch Part I : Perceptron

Contents

## Please view my work here!

Please checkout this Github Gist =>

To run the Colab just like I did, please click here🙂.

Special thanks to Nagesh’s tutorial which did an excellent job in explaining neural network in simple manners. Click to view part I of his tutorial series.

Recall the formula: $$w = w - \alpha (\frac{\partial F}{\partial w})$$
$\frac{\partial F}{\partial w}$ is the gradient of the loss F with respect to the weight w, which, to put simply, translates to rate change of the loss with the change of the weight and in 2-D term the slope. If the loss increases as weight increases, $\frac{\partial F}{\partial w}$ would be a positive value which is then subtracted from w by the amount $\alpha (\frac{\partial F}{\partial w})$, to go in the opposite direction of the gradient. The same applies if the loss decreases as weight increases: it means $\frac{\partial F}{\partial w}$ is a negative value which we in effect (minus a negative value) add the amount $\alpha (\frac{\partial F}{\partial w})$ to w. Either way, the goal is to push w in the direction that causes most rapid decline in F, the cost function we’re interested in minimizing.