13  Study definition

13.1 Reproducibile dummy data

• To ensure the dummy data system generates exactly the same data every time you run it, set a random number generator seed at the top of the study_definition.py file

  import numpy as np
  # Change this number to one for which your scripts 
  # successfully run on the dummy data
  np.random.seed(123456)

13.2 File formats

• Use .feather files for outputs from the cohortextractor, so specify an action in your project.yaml as follows

  generate_study_population:
    run: cohortextractor:latest generate_cohort --study-definition study_definition --output-format feather
   needs: 
    - design
    outputs:
      highly_sensitive:
        cohort: output/input.feather

• Use the arrow package to read .feather files into R

  arrow::read_feather(file = file.path("output", "input.feather"))

  • The col_select argument can be used to read in just the columns you need

• Start each project with a preprocessing action that formats .feather files and outputs (gzipped) .rds files which can be saved with readr::write_rds()

  readr::write_rds(object, 
                   file.path("output", "mydata.rds"), 
                   compress = "gz")