Project idea · ungraded

Gradient-Descent Visualiser

Build a tool that animates gradient descent on functions you choose — a bowl, a long narrow valley, a function with two minima — so you can see how the learning rate, the starting point and the shape of the function change the path.

You will use

Milestones

Build the smallest working version first. Ticks are saved in this browser only.

01Write gradient descent for one variable and print the sequence of points for f(x) = (x - 3)².
02Extend it to two variables using numerical gradients, for any function you pass in.
03Plot contour lines of f(x, y) with matplotlib and draw the path on top.
04Compare learning rates that converge, oscillate and diverge on the same function.
05Show a function with two minima where the starting point decides which one is found.

Stretch goals

Add momentum and compare its path in a narrow valley.
Animate the path with matplotlib's animation module and save a GIF.
Let the user type a formula and parse it safely.

Data

No data: the functions are your own. See matplotlib's animation guide.

Run it yourself

pip install numpy matplotlib
python descent.py