... | ... | @@ -87,6 +87,12 @@ Versions |
|
|
|
|
|
# ASR
|
|
|
|
|
|
ASR learning time and model selection
|
|
|
The selection of the learning model is based of segments of time that have a high AMICA log likelihood values.
|
|
|
ASR learn time ticks through each of the data points of the log likelihood and looks at the Lt values. By default for the time to be considered the one second block must have 80% of its points above them mean. This percentage can be adjusted, but its intent is prevent time segments where the model may be bad for part but the other half is very strong. If the following second has enough values above your critical mean then it is considered viable for the learning time. By progressing through each data point, sections of time will only be flagged for time sections for 1 sec or longer, but is not limited to 1 second intervals.
|
|
|
|
|
|
Viable time is then sorted so that the highest blocks log likelihood time are added first to the total time.Segments are continued to be added until the total time value defaulted at 1 minute is filled. This EEG data is added up and then sent to the *clear_asr* function.
|
|
|
|
|
|
| Variable | Default in Config File | Description |
|
|
|
|--------------------|-------------------------------------------|-------------|
|
|
|
| [in_path] | analysis/data/2_preproc | Relative path to input data files assuming cd = work_path |
|
... | ... | |