Skip to content

BasicFeatGenerator

This feature generator is used to create features for most common signals with 1 time channel and for a specific value channel (for example lab test like Hemoglobin). This feature calculate some stats in time window. It can be last value, mean value, slope, last_time in days of last test, etc. Here is an example block to define this feature generator - you should put this under "model_actions" element and after all the rep_processors and before feature_processors: 

1
2
3
4
{
    "fg_type": "basic",
    //ELEMENT initialization of BasicFeatGenerator 
}
  Full example on numeric values: 
1
2
3
4
5
6
7
8
{
    "fg_type": "basic",
    "type": [ "last", "max", "min", "avg", "slope", "last_time", "last2" ],
    "window: [ "win_from=0;win_to=365" ], //defines time window. I used list in here to be able to use this feature generator on multiple features. we can also use seperatly "win_from":"0", "win_to":"365"
    "signal": [ "Hemoglobin", "WBC", "Glucose" ], //The signal to operate on
    "val_channel":"0", //if signal has more than 1 channel (for example BP) we can specify on each value channel to work. Deafult is 0
    "tags": "labs,numeric,need_imputer" //we can specify "tags" for this group of feature to later refer all of them. For example, do imputations for all features with "need_imputer" tag
}
Categorical example: 
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
{
    "fg_type": "basic",
    "type": "category_set" , //The operator, can also be category_set_count to store the count
    "window: [ "win_from=0;win_to=365" ], //defines time window. I used list in here to be able to use this feature generator on multiple features. we can also use seperatly "win_from":"0", "win_to":"365"
    "signal": "Drug" , //The signal to operate on
    "val_channel":"0", //if signal has more than 1 channel (for example BP) we can specify on each value channel to work. Deafult is 0
    "sets": "ATC_A10B_A02,ATC_A10B_A01", //defined code or list of codes (comma separated) to construct the "set". You can also refer to codes in file with "comma_rel:$FILE_PATH_OF_CODES"
    "in_set_name": "Diabetes_drug", //Optional argument to give name to this set. Otherwise a default of concatinating codes from "sets" will be created as the name of the feature
    "tags": "categorical" //we can specify "tags" for this group of feature to later refer all of them. For example, do imputations for all features with "need_imputer" tag
}
    For full list of arguments, please refer to: https://Medial-EarlySign.github.io/MR_LIBS/classBasicFeatGenerator.html   For full list of "type" params: https://Medial-EarlySign.github.io/MR_LIBS/FeatureGenerator_8h.html Here are some common operators/types:

  • "last" - take last value in the time window
  • "max" - take maximal value
  • "last_time" - take last time in days of the test in time window (in days)
  • "category_set" - for categorical signal. boolean result. Will test if has value part of "sets" parameter in the defined time window
  • "category_set_count" - for categorical signal. numeric result. Will count how many times found a value that is part of "sets" parameter in the defined time window   For full json model format refer to MedModel json format