Pick Activity: a neural network you can train

2 inputs → 8 hidden neurons → 5 outputs. It starts with random weights, so the predictions are garbage. Press Train and watch it learn the rules from examples.

Inputs

Temperature 72 °F Chance of rain 20 %

Network says

The if/else rules say:

The rules (what the network has to discover)

Temperature	Rain %	Activity
≥ 80 °F	< 50%	play volleyball
≥ 80 °F	≥ 50%	see a movie
32 – 79 °F	< 50%	go running
32 – 79 °F	≥ 50%	try yoga
< 32 °F	any	read a book

These are the same rules as the if/else program in Part 1, which checks them one line at a time.

Training

Untrained — weights are random

Epoch

Loss

—

Accuracy

—

Show the 56 weights (they change as it trains)

W1 — inputs (rows: temp, rain) → 8 hidden neurons (columns)

W2 — 8 hidden neurons (rows) → 5 activities (columns)

Show the training data (2000 example days)

Each row is one example: the inputs and the answer the if/else rules gave. The network sees temp ÷ 100 and rain ÷ 100.

#	Temp	Rain	Activity

The network never sees the rules table. It only sees 2000 example days that were labeled by those rules. Try sliding to exactly 80°, exactly 32°, or exactly 50% rain after training.