 this is the alpha application, this is actually what we have the description here for so sorry when working on that one should have had a different description this is the one we plan to release shortly for you to play with can then now set your noise levels, that's the noise level at default you can obviously still play the game as in the original tool, oops not doing well am I not paying attention that's the normal ECG so I should get 10 points for not clicking on it and this one missing a bit so I do click on it and get another 10 points so I got one hit one miss and the current bot which we changed the parameters of is doing badly it's essentially clicking on everything so I can win the bot except that yes we can artificially tune it up and it will work much better ideal we're trying not to use any magic numbers in the algorithm so it's it's all the thresholds and everything are just there as if they were provided by yeah so those are the rules so this is the fuzzy logic analysis we're looking before we're looking at amplitude positive peak negative peak abnormality and frequency now we get rid of frequency looking at the sharpness of the up peak so assuming the up peak is the highest positive peak and then we're going I think half a way measuring the distance between you know left and right coordinates in it that that works fine then we normalize it so we'll get a 0.4 sharpness for like a regular thing and a 0.1 for a less sharp up peak so that's working fine except there's still something wrong so for this signal made the decision that it's an abnormal ECG because the abnormalities score didn't go low enough and that's a problem and this is because we haven't done the membership functions correctly so for example with amplitude we have a low medium and high because we're more normalizing it we are not normalizing anymore so the amplitude should change the maximum amplitude is about the yeah which is so before three five six so we have to fix the code question is should we do today as a false alarm clicking on a normal normal ECG well the question the question always is I can't call myself question is if the code the of the ports can fix this for us try core pilot normally in this let's start a new one is this couple of things so our sharpness is calibrated now the problem with raw amplitude is there a calibration for the raw amplitude do we normalize that value or do we use the raw value if so it is not from 0 to 1 we might need to fix the code the maximum value is about 360 that's looking at the entire python script yeah it's adjusting normalization dividing by the maximum value and say quickly amplitude universe mentioning amplitude like 37 times amplitude yeah when we have the raw one yeah should we normalize something to function a normalize using a max value should we make it like 400 or something yeah we don't like magic numbers that's a quick fix um okay let's just give me zero all the time we should we try not looking at amplitude at all now it's one that was a normal I know it wasn't normal yeah that's normal the raw amplitude is always one a what should I divide it by was it 350 or something wasn't it why is this giving me zero we print this value out clear 0.8 so yeah that's working perfectly it's just not showing a decimal places space raw amplitude is not showing decimal places it's finding what to look at it's taking forever sorry I was muted I can leave all the comments so it knows what we were doing and we have this image as an example this so we have a code now this code should have processed the waveform in the image as normal ACG waveform it had the good number of positive and negative peaks the amplitude should have been labeled as normal can we double check that I think that might be what's what the problem is sharpness it should have been okay so the abnormalities score should have been low enough for the algorithm to make a decision that the waveform is normal yep can we go over each membership function and each rule one at a time specifically the one to do with amplitude because that's what we changed recently the calculation of it just realized that the frequency actually was detecting the difference between normal and one of those examples where the r peak is shifted forward in time because that was producing a frequency of five and now we do not have a variable to detect it we don't have a feature that we can use to detect that variation yeah every time you fix something you break something else isn't it yeah so we already have this rule should be working okay yeah the amplitude needs to be tuned up I think I know what the problem might be the amplitude scaling let's try 500 for a sec we'll get the amplitude of 0.6 then 0.7 so 0.7 should be in the normal range which is not quite true yeah this meant to be normal yeah we can manually tune it up that's 0.6 0.3 abnormal 0.6 it's classifying that one it's normal I don't know why the rules should just make sense it's always one of those things that should have just designed you know once and not touch ever again should have just worked yeah all right well that rule now needs to I'm not sure about this one might get a might rewrite it without looking at the amplitude it don't include amplitude just keep it simple yeah so if positive peaks uh too few or too many the abnormality is likely yeah understand now it kind of makes sense I understand now why that first one gave a miss because it was abnormal this one is wrong by the way why there's two negative peaks that's abnormal should have only one negative peak yeah there's something wrong with the negative peak detector so that one is correct and this one wasn't that's a case of one negative peak when there is a should be zero yeah this one can we check the negative peaks um the counter the detector is uh incorrect it the I don't know can you explain why in the first two images it's detecting two negative peaks and the last image is detecting one in all three cases there should be just one negative peak example data in the image provided should have given only one negative peak instead it's giving two can you look at the code and explain why and how to fix doing filtering somewhere already because last time I was saying if the bot has access to filtered data the human should as well so might like overlay the filtered a bit over the black one on the top yes we expected to be smaller amplitude is there any filtering in this code which lines are filtering in this python code it's because it's too long yeah might be way too long not doing frequency might be a problem okay can you summarize the whole session the whole code that you have access to all the modifications that we made can you also touch on the difference between uh peak sharpness calculation and frequency analysis also can you check the code for any filtering that we are doing I already checked we do not use field functions but I think there might be a different more simple filter can you find that and we might be finishing now so can you give an overall summary of how this application can be useful and if there is anything similar in the wild yeah we really broke the fuzzy logic detector it doesn't work at the moment we'll have to fix it next time to go check out binarykills.com and don't forget to provide your feedback see you next time bye