Reduce function execution time in spike neural network

Question

I'm trying to generate spikes from image for spike neural networks. I do this with rate coding, where the probability of generating a spike corresponds to the pixel value. I test my function with 28x28 pixel image and the execution time was ~ 0.65 sec. If I'll generate spike using MNIST datset, it is gonna take about 10 hours to do, what is quite bad. How can I boost my function?

Here's the code I wrote:

def spike_generator(image_list, total_time, delay_time, frequency, seed=42):

    '''
    Function generates spike patterns based on the input data array.
    Probability of spiking corresponds to the pixel value.

    Parameters:

    image_list: list of numpy arrays containing input data for each neuron
    total_time (float): total duration for spike generation in seconds
    delay_time (float): time delay between spikes for each neuron in seconds
    frequency (int): frequency of spike generation in Hz
    seed (Optional(int), default = 42): a numerical value that
    generates a new set or repeats pseudo-random numbers

    Returns:

    generated_spikes: numpy array containing spike patterns
    for each neuron over the specified total_time duration
    '''

    np.random.seed(seed)
    generated_spikes = np.zeros((image_list.shape[0], image_list.shape[1], total_time * frequency))

    for z in range(image_list.shape[0]):
        data_norm = (image_list[z] - image_list[z].min()) / (image_list[z].max() - image_list[z].min())
        for i in range(image_list.shape[1]):
            while f <= total_time * frequency:
                if np.random.choice([0, 1], p=[1 - data_norm[i], data_norm[i]]) == 1:
                    generated_spikes[z][i][f] = 1
                    f = f + mt.ceil(frequency * delay_time)

    return generated_spikes

I've tried to use numpy.vectorize, but this didn't reduce execution time.