fixed small error in the global threshold calculation

[bnestor]

Addresses #219

The number of positive calls had to be greater than the min_num_positive_calls_threshold to trigger a global prediction

[pastorep]

I'm a bit confused about the impact of this bug (which I think is secondary to the issue you raise here): #219 (comment).

Semantically, I understand that "minimum" suggests inclusivity. Meaning, if min positive call == 1 and I have one positive call, I should meet that threshold.

In the referenced thread, you mention that you had 15 seconds of high SNR (signal-noise ratio) in a global 15 second timeframe. In that case, the model detected only one individual positive call for the full 15 seconds and didn't meet the "true minimum" of 2 (i.e. no second call because the first call took full 15 seconds)? Is this the correct understanding?

[bnestor]

My model uses 15-second subsamples of the 1-minute audio files. The local time-frame infers on 15 second chunks instead of 2-second chunks in the fastai model. So in my global model, I only need 1 of the local time-frames to be positive.

local_predictions_15s_windows_15s_hoplength=[0,1,0,0] # should be a global prediction of 1, but is currently 0
local_predictions_2s_windows_2s_hoplength=[0,0,0,0,0,0,0,0,1,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0] # practically no impact on the current model behaviour, especially when the hop length is less than the window length.