SMNIST: unexpected behavior of "duplicate" flag

[mhastu]

In the SMNIST dataset, per default, the "duplicate" flag is set, which means two spikes are emitted per threshold crossing. This, however, silently also doubles all the spike timings, i.e. the spikes are fed at half the speed into a model, which is seen in the following snippet:

        # If we should duplicate each spike
        if self.duplicate:
            # Repeat spike indices
            spike_idx = np.repeat(spike_idx, 2)

            # Double spike times
            double_spike_time = spike_time * 2

            # Interleave
            spike_time = np.empty(2 * double_spike_time.shape[0], dtype=np.int64)
            spike_time[0::2] = double_spike_time
            spike_time[1::2] = double_spike_time + 1

We might want to

• half dt and add half dt to the original spike timings and document this in the class description string
• keep the behavior as it is and document the spike timing doubling in the class description string
• default the "duplicate" flag to False, independently of the above suggestions.