Teach your laptop your hand signs with two classic models: logistic regression and a decision tree.
The model only acts when its top probability is above the cutoff. Otherwise it says "not sure". This model and cutoff are used everywhere: live guess, test, game.
Hold the button (or keys 1 2 3) while you show the sign. Move your hand around a bit: closer, further, tilted. Aim for 100+ examples per sign. You can rename the signs.
Pick what the models see, then train. The tree retrains as you move its sliders.
Record a few seconds per sign in a different way: other hand, other distance, stand up, ask someone else. These examples are never used for training, only for grading.
Your model only knows your hand. Delete your own test data, load a teammate's file, and see what happens.
Everything here is computed on the test data with the model chosen under the camera.
Uses the model and cutoff chosen under the camera. Throw your sign on "shoot!". The model watches for half a second and takes the sign it saw most. If it is "not sure" most of that time, you lose the round. Signs must be named Rock, Paper, Scissors in this order.
The tree is a list of if-then rules you can read. Logistic regression is one weight per feature per sign.
Same models in Python: download your data as CSV (step 1), then open hand_signs_starter.ipynb in Colab or Databricks.