Benchmarking - CCDEF - Fidelity check sign error

[briangow]

I get an error like the one in the example below when running the CCDEF format code for benchmarking against any file from https://physionet.org/content/charisdb/1.0.0/ . The values from the written file seem close to the input data but the sign is wrong.

./waveform_benchmark.py -r "/home/briangow/chorus/data/waveform_benchmark/waveforms/charisdb/1.0.0/charis1.hea" -f "waveform_benchmark.formats.ccdef.CCDEF_Compressed"
________________________________________________________________
Format: waveform_benchmark.formats.ccdef.CCDEF_Compressed
         (CCDEF compressed format)
Record: /home/briangow/chorus/data/waveform_benchmark/waveforms/charisdb/1.0.0/charis1
         244797 seconds x 3 channels
         36719553 timepoints, 36719553 samples (100.0%)
________________________________________________________________
Fidelity check:

Chunk		 Numeric Samples		  NaN Samples
	# Errors  /  Total	  % Eq  	NaN Values Match
Signal: ABP
  0  	         2/  12239851	100.000		     Y (0)      
Subset of unuequal numeric data from input:
[387.2814 396.7168]
Subset of unuequal numeric data from formatted file:
[-375.18178539 -365.74639629]
(Gain: 85.953)
Signal: ECG
  0  	       246/  12239851	99.998		     Y (0)      
Subset of unuequal numeric data from input:
[5.4534044 5.4534044 5.4534044 5.4534044 5.4534044 5.4534044 5.4534044
 5.4534044 5.4534044 5.4534044]
Subset of unuequal numeric data from formatted file:
[-5.32284526 -5.32284526 -5.32284526 -5.32284526 -5.32284526 -5.32284526
 -5.32284526 -5.32284526 -5.32284526 -5.32284526]
(Gain: 6081.5219)
Signal: ICP
  0  	         3/  12239851	100.000		     Y (0)      
Subset of unuequal numeric data from input:
[357.78427 346.12726 369.76804]
Subset of unuequal numeric data from formatted file:
[-332.95249498 -344.60952124 -320.96873472]
(Gain: 94.8784)

[briangow]

This error is being introduced at:

chorus_waveform/waveform_benchmark/formats/ccdef.py

Line 55 in 576e56d

sig_samples[start:end] = np.round(cursamples*chunk["gain"])

when a sample value is outside the np.short (16 bit) range.

@wa6gz , assuming you'd like to stick with np.short to minimize the file size, could you adjust the baseline to stay within the 16 bit range?

[wa6gz]

Sounds good @briangow, I'll add a baseline and integrate with gain changes in #65.

[briangow]

Thanks @wa6gz !