features output of guanlab_model(), are 1s or 0s

[itismeghasyam]

library(mhealthtools)

### Load data file
data("sensor_data")
dat <- mhealthtools::sensor_data

### flatten data to the format needed for mHealthTools
flatten_data <- function(dat, metric) {
  dat <- dat %>% 
    select(timestamp, metric) %>% 
    jsonlite::flatten()
  names(dat) <- c("t", "x", "y", "z")
  return(as_tibble(dat))
}

### Get the formatted accelerometer and gyroscope data to use in testing below
datAccel <- flatten_data(dat,'userAcceleration')

#### Test GuanLab model
guanlab_model <- function(sensor_data, models = load_guanlab_model) {
  if (is.function(models)) { # load weights from file
    models <- models()
  }
  features <- purrr::map_dfc(models, function(model) {
    padding <- matrix(0, 4000 - dim(sensor_data)[1], 3)
    standard_sensor_data <- sensor_data %>%
      dplyr::select(-t) %>% 
      purrr::map(~ (. - mean(.)) / sd(.)) %>% 
      unlist() %>% 
      matrix(ncol = 3, byrow = F)
    padded_sensor_data <- standard_sensor_data %>% 
      rbind(padding) %>% 
      keras::array_reshape(c(1, 4000, 3))
    proba <- model$predict_proba(padded_sensor_data)
    return(proba)
  })
  return(features)
}

#### feature output
a <- mhealthtools:::guanlab_model(sensor_data = datAccel)

 >a
# A tibble: 1 x 10
     V1    V2    V3    V4    V5    V6    V7    V8    V9   V10
  <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1     0     0     1     1     0     0     0     1     0     0

When I run this a is just a vector of 0s and 1s, shouldn't it be probabilities instead of class predictions?
I think predict_proba is just giving out 1s and 0s as output, instead of the actual probability

[philerooski]

We discussed this last month. Closing because this is superseded by #37