Question about Perceptron.ipynb

[how2051]

Hi everyone~

I've been reading these several file in section 3 -- Perceptron, but I still got confused.

Here's my confusion:

In the Perceptron.ipynb file, we first generate some random data that have two dimensions/features, and we use 80% of them as trainning data, and 20% of them as test data

and then comes to this train function, where I started to get confused:

in this func, we first get num_dims, which as postitive example 's column, stands for the number of the feature, which will be 2
and our weight will be a vector of 1 col and num_dims rows, so it going to match the feature

in the for loop, we pick up a random example, and dot product it with weight.

now, since the weight was init to zero, HOW would the result of dot product be non-zero value ?
wouldn't that be always zero ?

        z = np.dot(pos, weights)   
        if z < 0: # positive example was classified as negative
            weights = weights + pos.reshape(weights.shape)

        z  = np.dot(neg, weights)
        if z >= 0: # negative example was classified as positive
            weights = weights - neg.reshape(weights.shape)

[how2051]

idk if it's because I'm not familiar with python, or just I miss something, I just cant understand.
Am I right so far? Or are there any misunderstanding I make ?
It would be very appreciated if someone can explain me this, thanks a lot