rixpress translation of Yanai and Lercher 2020

This pipeline is similar to the example found in python_r in that it mixes both R and Python code and ultimately outputs an HTML document.

This is inspired by the code found in this blog post. Thank you to Isabella Velásquez for the suggestion!

You can find the original paper here.

Source code is here. The pipeline definition is in gen-pipeline.R and the environment definition in gen-env.R.

Let’s start by looking at the original image:

Original

Get the coordinates using Python by defining this derivation:

d0 <- rxp_py_file(
  name = gorilla_pixels,
  path = 'md_source/gorilla/gorilla-waving-cartoon-black-white-outline-clipart-914.jpg',
  read_function = "read_image",
  additional_file = "functions/functions.py"
)

This is the of the read_image function:

import numpy as np
from PIL import Image

def read_image(x):
    im = Image.open(x)
    pixels = np.asarray(im)
    return pixels

We then define a threshold and get the remaining coordinates:

d1 <- rxp_py(
  name = threshold_level,
  py_expr = "50"
)

d2 <- rxp_py(
  name = py_coords,
  py_expr = "numpy.column_stack(numpy.where(pixels < threshold_level))"
)

We can then convert py_coords to an R object:

d3 <- rxp_py2r(
  name = raw_coords,
  expr = py_coords
)

Then we clean them:

d4 <- rxp_r(
  name = coords,
  expr = clean_coords(raw_coords),
  additional_files = "functions.R"
)

This is the R function used to clean the coordinates:

clean_coords <- function(coords) {
  as.data.frame(coords) %>%
    dplyr::sample_n(1768) %>%
    dplyr::mutate(bmi = V2 * 17 + 15, steps = 15000 - V1 * 15000 / max(V1)) %>%
    dplyr::mutate(
      randvar = rnorm(n(), mean = 0, sd = 10),
      randi = steps * (1 + randvar),
      gender = dplyr::case_when(randi < median(steps) ~ "Female", TRUE ~ "Male")
    )
}

Finally, we can look at the gender distribution:

rixpress::rxp_read("/nix/store/vrnzh9qaln0pv03i6qj3fkr5q3x901yj-gender_dist")

  gender   n
1 Female 929
2   Male 839

and some graphs:

rixpress::rxp_read("/nix/store/74j8q9z03llh9hg3d5d4wsmg6isbsv7m-plot1")

rixpress::rxp_read("/nix/store/s01aqbrzgsvpww4gwwwsirmsmlnc2k06-plot2")

The code to compute the gender distribution and graphs:

gender_distribution <- function(coords) {
  dplyr::count(coords, gender)
}

make_plot1 <- function(coords) {
  coords %>%
    ggplot(aes(x = bmi, y = steps)) +
    geom_point() +
    theme_void() +
    xlim(0, 15000)
}

make_plot2 <- function(coords) {
  coords %>%
    ggplot(aes(x = bmi, y = steps, color = gender)) +
    geom_point() +
    theme_void() +
    xlim(0, 15000)
}