Python vs. R in org-mode babel output graphics

I am trying to write up a report in org-mode. Read data from a csv file (single column three rows, floating numbers), compare in a bar chart, have the chart embedded in the report so that it can be exported to latex and then to pdf.

I am having difficulty understanding as in what I am doing in bar creation part of the python code because R_plot works fine, meaning chart is embedded into the report under the same org-mode :export :results :file setting.

What am I doing wrong with the python code? If I run the python code on interactive mode, it produces the chart with no trouble but for some reason py_comparison.png isn't saved when I run by cell-block.

#+NAME: R_plot
#+BEGIN_SRC R :exports both :results output graphics :file r_comparison.png
# graph in R
library("ggplot2")
performance <- read.csv("comparison.csv", header=FALSE)$V1
df <- data.frame(resource = c("1node1core", "1node8core", "2node8core"), performance = performance)
p <- ggplot(data = df, aes(x=resource, y=performance)) +
     geom_bar(stat="identity", fill="steelblue") + 
     theme_minimal() +
     ggtitle("Computation time (min) vs. Resource (type)")
p
#+END_SRC

#+NAME: python_plot
#+BEGIN_SRC python :exports both :results output graphics :file py_comparison.png
import matplotlib.pyplot as plt; plt.rcdefaults()
import matplotlib.pyplot as plt
import csv

objects = ['1node1core', '1node8core', '2node8core']
y_pos = list(range(0, len(objects)))

performance = []
with open('comparison.csv', newline='') as csvfile:
  reader = csv.reader(csvfile)
  for row in reader:
    f_row = float(row[0])
    performance.append(f_row)

plt.bar(y_pos, performance, align='center', alpha=0.5)
plt.xticks(y_pos, objects)
plt.ylabel('Time')
plt.title('Resource vs. Time')

plt.show()
#+END_SRC

Solution

I believe your header value for python code block is wrong. There is a difference between :results <file name> file and :file <file name>. According to the documentation (redacted for clarity):

:results file

There are four classes of :results header arguments. Each ‘src’ code block can take only one option per class. [...]

Collection [...]

value Default. Functional mode. Result is the value returned by the last statement in the ‘src’ code block. Languages like Python may require an explicit return statement in the ‘src’ code block. Usage example: :results value.

Type [...]

file Interpret as path to a file. Inserts a link to the file. Usage example: :results value file.

:file <file name>

An external :file that saves the results of execution of the code block. [...] Some languages, such as ‘R’, ‘dot’, ‘ditaa’, and ‘gnuplot’, automatically wrap the source code in additional boilerplate code. Such code wrapping helps recreate the output, especially graphics output, by executing just the :file contents.

In python the result of plt.show() (or savefig for that matter) is None, the image is only a side effect, so nothing gets inserted. In R it works because of the boilerplate wrapper mentioned above

So in python you need to save and return your image instead of showing it. Something like this should work:

#+NAME: python_plot
#+BEGIN_SRC python :results img.png file

import matplotlib.pyplot as plt
plt.plot(range(5))
plt.savefig('img.png')
return 'img.png'

#+END_SRC