A Neuron Is a Weighted Vote

A vector is an ordered list of values. Position matters because each position will be multiplied by its corresponding weight. Here [1, 1, 0] means a friend is going, the event is nearby, and rain is not forecast.

The computer sees numbers; your data contract supplies their meanings. Swapping rain and friend would change the decision while still producing a perfectly valid numerical calculation.

Store a three-item list under features and label its output with the intended order.

The weights are 0.6, 0.3, and -0.4. Positive contributions support activation; the negative rain contribution opposes it. The threshold 0.5 is the bar the total must reach.

This is a manually designed threshold neuron. We have not trained these numbers or claimed that hard thresholding is how every modern network unit works.

Create decision with a list-of-weights value and threshold 0.5. Keep three weights for three input features.

The products are 1 × 0.6, 1 × 0.3, and 0 × -0.4. They sum to 0.9. The third feature contributes zero because no rain was encoded, even though its weight is negative.

A weighted sum is a number, not yet the final yes/no decision. Keep that intermediate value visible before applying the threshold.

Put the input list in weighted total of decision, then put that value in labelled output.

The neuron fires when its total is greater than or equal to 0.5. Since 0.9 reaches that threshold, the result is True. Equality is included by this implementation.

The activation turns the weighted total into a particular kind of output. A hard threshold returns a Boolean, while other activations can preserve a range of numerical values.

Print both total and fires for exactly the same input list.

With friend present and rain absent, the total is 0.6. Turning rain on contributes -0.4, reducing the total to 0.2. That falls below 0.5, so this manually chosen decision no longer fires.

The sign of the contribution comes from both the input and weight. Here the input is positive one, yet its contribution is negative because the weight is negative.

Compare the two lists while keeping all weights and the threshold fixed.

Comparing sum with threshold is equivalent to subtracting the threshold and comparing with zero. For a threshold of 0.5, the bias in this form is -0.5. The total 0.9 becomes 0.4 after adding that bias.

A neural-network unit commonly computes a weighted sum plus bias before an activation. This algebra connects the inspectable teaching block to that more general notation.

Store the weighted total, subtract 0.5, and compare the result with zero.

These four cases use one fixed neuron. Their totals are 0.3, 0.6, -0.1, and -0.4. Only 0.6 reaches the threshold. A loop lets you inspect each case without reconstructing the weights between them.

Testing a set of deliberate cases is stronger than showing only the one positive example that motivated your weights. You can see which features help, oppose, or contribute nothing.

Build the list of input lists, then print input, total, and fires inside for each.

This final stack stores the activation before choosing a message. The weighted sum, activation, and response remain separate operations. In an actual learned model, training would choose suitable parameters from examples and a loss.

Practical neural networks often use ReLU, sigmoid, or other activations compatible with their learning procedure, rather than training a raw hard threshold by ordinary backpropagation. The small block is a transparent arithmetic model, not a complete replica of all neural units.

Choose your fictional input list, inspect the total, and read the activation-driven response. Compare your construction with the collapsed reference only after tracing it.