Add example for Python/R interoperability using mass univariate t-test by aman-coder03 · Pull Request #13729 · mne-tools/mne-python

aman-coder03 · 2026-03-09T15:17:42Z

Reference issue (if any)

What does this implement/fix?

adds a tutorial example that runs a 2 sample mass univariate t-test on Epochs data,.., first in python using scipy.stats then the same test in R via rpy2 showing that both give identical results. The goal is to show the users how to get data into R, run stats there, and then bring the results back into Python, as starting point for more complex R based workflows

Additional information

follows the approach suggested by @larsoner in the issue ...keeping it simple and focused on one clear use case rather than trying to build general purpose utility

aman-coder03 · 2026-03-09T17:09:43Z

windows 3.13 pip pre failure is unrelated to this PR, it's a missing nightly wheel for statsmodels>=0.15.0.dev697 on the scientific python nightly wheels index, which affects other PRs too

larsoner · 2026-03-09T17:11:24Z

Being fixed in #13728

larsoner · 2026-03-13T15:58:45Z

CircleCI failure is real

I think you need to add rpy2 to the doc dependency group in pyproject.toml

for more information, see https://pre-commit.ci

aman-coder03 · 2026-03-13T16:59:08Z

@larsoner the failure is because rpy2 requires R to be installed at build time, which isn't available in the CI environment. Would wrapping the rpy2 imports in a try/except to skip the R section gracefully when R isn't present be the right approach here?

larsoner · 2026-03-13T17:50:39Z

Can you try updating the CI to have R installed? Would be doable with an apt command

for more information, see https://pre-commit.ci

aman-coder03 · 2026-03-14T09:13:44Z

towncrier fragments errors about "Famous Raj Bhat" are pre existing and unrelated to this PR.. @larsoner

Famous077 · 2026-03-16T14:09:58Z

towncrier fragments errors about "Famous Raj Bhat" are pre existing and unrelated to this PR.. @larsoner

Hi @aman-coder03 , Can you specify where you are getting an error. Would Love to co-ordinate with you to fix the issue.

larsoner · 2026-03-16T19:04:48Z

Hi @aman-coder03 , Can you specify where you are getting an error. Would Love to co-ordinate with you to fix the issue.

No need, I fixed it in main and then merged main into this branch

aman-coder03 · 2026-03-16T19:51:06Z

thanks @larsoner !!

larsoner · 2026-03-16T20:00:57Z

tools/circleci_dependencies.sh

@@ -1,5 +1,7 @@
 #!/bin/bash -ef

+sudo apt-get install -y r-base libtirpc-dev


This isn't the best place, tools/circleci_bash_env.sh would make more sense

larsoner

Example looks good

https://output.circle-artifacts.com/output/job/05c66585-d8d4-43fe-b5d3-bdb6fa583fc1/artifacts/0/html/auto_examples/preprocessing/r_interop.html#sphx-glr-auto-examples-preprocessing-r-interop-py

@drammock feel free to merge if you're happy, I defer to your (far) superior R experience!

drammock

In the long run, I think it would be better / more interesting to showcase an example of something where R/rpy2 actually makes it easier to achieve the desired result (i.e., demonstrate something that is hard to do in Python but easy in R). But for now, however, let's stick with the t-test (subject to the modifications I propose in the below comments)

drammock · 2026-03-17T15:07:26Z

examples/preprocessing/r_interop.py

file is in the wrong folder. This should be in stats folder, not preprocessing

drammock · 2026-03-17T15:10:38Z

doc/changes/dev/13729.newfeature.rst

@@ -0,0 +1 @@
+Add example showing how to run a mass-univariate t-test in both Python and R using ``rpy2``, demonstrating Python–R interoperability with :class:`~mne.Epochs` data by `Aman Srivastava`_. (:gh:`13729`)


Suggested change

Add example showing how to run a mass-univariate t-test in both Python and R using ``rpy2``, demonstrating Python–R interoperability with :class:`~mne.Epochs` data by `Aman Srivastava`_. (:gh:`13729`)

Add example showing how to interoperate with R using ``rpy2``, by `Aman Srivastava`_.

PR number will get added automatically, and the details (epochs, t-test) are not really the point.

examples/preprocessing/r_interop.py

drammock · 2026-03-17T15:22:20Z

examples/preprocessing/r_interop.py

+# We average over channels and time to get one value per epoch, then run
+# a 2-sample t-test comparing the two conditions.


this is... not something anyone would ever do in practice. What we show can be much simpler than what most real neuroscience studies do, but it should at least be plausible. Probably the shortest path to something plausible would be:

pick a few sensors as an "ROI" (ones that should differ between auditory L vs R conditions)

pick a typical ERP time window

average over just those channels and those times

do the t-test (1 observation per epoch)

Probably worthwhile to do a plot_compare_evokeds or similar, as a way to motivate what the stats are testing.

drammock · 2026-03-17T15:30:09Z

examples/preprocessing/r_interop.py

+with localconverter(default_converter + numpy2ri.converter):
+    r_left = ro.FloatVector(epochs_left)
+    r_right = ro.FloatVector(epochs_right)
+
+r_ttest = ro.r["t.test"]
+result = r_ttest(r_left, r_right, **{"var.equal": True})
+
+t_r = float(result.rx2("statistic")[0])
+p_r = float(result.rx2("p.value")[0])


these steps are the important "new learning" from this, so there should be more explanation/comments here. E.g., things like

rpy2.robjects (imported here as ro) has an attribute r that provides access to the R namespace, through a dictionary-like interface

Unlike Python, R functions can have parameter names that contain a ., so you may need to pass function arguments as a dict and unpack them with **

what is FloatVector and why do we need it

what is the rx2 attribute of the result, and why do we need the [0] after it

aman-coder03 · 2026-03-18T06:55:37Z

moved example to stats, added ROI + evoked plot, and explained the rpy2 API in more detail.

drammock · 2026-03-20T20:20:44Z

examples/stats/r_interop.py

+# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+#
+# We first plot the evoked responses to motivate the statistical test.
+# Auditory left vs right stimuli should differ over lateral temporal sensors.


EEG059 is not really lateral/temporal. Use MEG 1323

drammock · 2026-03-20T20:24:57Z

examples/stats/r_interop.py

+# and a typical N1 time window (80–120 ms). This gives one value per epoch,
+# which is a plausible neuroscience analysis.
+
+roi_channels = ["EEG 059", "EEG 058", "EEG 057"]


Suggested change

roi_channels = ["EEG 059", "EEG 058", "EEG 057"]

roi_channels = ["MEG 1323"]

drammock · 2026-03-20T20:25:01Z

examples/stats/r_interop.py

+    .crop(tmin_roi, tmax_roi)
+    .get_data(picks=roi_channels)
+    .mean(axis=(1, 2))
+)


unnecessary copying. something this should be fine:

epochs.crop(tmin, tmax).pick(roi_chs) epo_left = epochs["auditory/left"].get_data().mean(axis=(1, 2)) epo_right = epochs["auditory/right"].get_data().mean(axis=(1, 2))

drammock · 2026-03-20T20:29:42Z

examples/stats/r_interop.py

+
+mne.viz.plot_compare_evokeds(
+    {"auditory/left": evoked_left, "auditory/right": evoked_right},
+    picks="EEG 059",


Suggested change

picks="EEG 059",

picks="MEG 1323",

drammock · 2026-03-20T20:30:33Z

examples/stats/r_interop.py

+.. _r-interop:
+
+==============================================
+Mass-univariate t-test: Python and R compared


Let's make the point of the tutorial clearer in the title

Suggested change

Mass-univariate t-test: Python and R compared

Integrating with R via rpy2

aman-coder03 requested review from agramfort, drammock and larsoner as code owners March 9, 2026 15:17

add mass univariate t-test python/R interoperability example

fd55fd3

aman-coder03 force-pushed the doc-r-interop branch from 55cfb78 to fd55fd3 Compare March 9, 2026 20:44

sharifhsn mentioned this pull request Mar 10, 2026

PERF: Speed up permutation cluster tests via Numba union-find + compact graph #13731

Open

aman-coder03 and others added 3 commits March 13, 2026 22:11

adding rpy2 to doc dependencies

7c0da3b

[pre-commit.ci] auto fixes from pre-commit.com hooks

8400d1c

for more information, see https://pre-commit.ci

ci: retrigger

09b5253

aman-coder03 and others added 6 commits March 13, 2026 23:29

install R

a8df4dc

add libtirpc-dev for rpy2 build

2d3b438

fix rpy2 cross reference warnings

88af35f

[pre-commit.ci] auto fixes from pre-commit.com hooks

bc4ab04

for more information, see https://pre-commit.ci

fix warnings

2218a09

fix author and rpy2 reference warnings

7bbe0c3

Merge branch 'main' into doc-r-interop

b9e6750

Merge branch 'main' into doc-r-interop

ac6cbcd

larsoner reviewed Mar 16, 2026

View reviewed changes

aman-coder03 and others added 2 commits March 17, 2026 09:20

move r-base and libtirpc-dev to circleci_bash_env.sh

fb56f6f

Merge branch 'main' into doc-r-interop

c8c012d

larsoner approved these changes Mar 17, 2026

View reviewed changes

drammock requested changes Mar 17, 2026

View reviewed changes

aman-coder03 and others added 3 commits March 18, 2026 00:01

move r_interop to stats and expand rpy2

2478bd2

Merge branch 'main' into doc-r-interop

7af2aab

fix rpy2 cross-reference warnings

2b02633

aman-coder03 requested a review from drammock March 18, 2026 11:52

drammock requested changes Mar 20, 2026

View reviewed changes

		@@ -1,5 +1,7 @@
		#!/bin/bash -ef

		sudo apt-get install -y r-base libtirpc-dev

		@@ -0,0 +1 @@
		Add example showing how to run a mass-univariate t-test in both Python and R using ``rpy2``, demonstrating Python–R interoperability with :class:`~mne.Epochs` data by `Aman Srivastava`_. (:gh:`13729`) No newline at end of file

	Add example showing how to run a mass-univariate t-test in both Python and R using ``rpy2``, demonstrating Python–R interoperability with :class:`~mne.Epochs` data by `Aman Srivastava`_. (:gh:`13729`)
	Add example showing how to interoperate with R using ``rpy2``, by `Aman Srivastava`_.

		# We average over channels and time to get one value per epoch, then run
		# a 2-sample t-test comparing the two conditions.

	roi_channels = ["EEG 059", "EEG 058", "EEG 057"]
	roi_channels = ["MEG 1323"]

	Mass-univariate t-test: Python and R compared
	Integrating with R via rpy2

Uh oh!

Conversation

aman-coder03 commented Mar 9, 2026

Reference issue (if any)

What does this implement/fix?

Additional information

Uh oh!

aman-coder03 commented Mar 9, 2026

Uh oh!

larsoner commented Mar 9, 2026

Uh oh!

larsoner commented Mar 13, 2026

Uh oh!

aman-coder03 commented Mar 13, 2026

Uh oh!

larsoner commented Mar 13, 2026

Uh oh!

aman-coder03 commented Mar 14, 2026

Uh oh!

Famous077 commented Mar 16, 2026

Uh oh!

larsoner commented Mar 16, 2026

Uh oh!

aman-coder03 commented Mar 16, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

larsoner left a comment

Choose a reason for hiding this comment

Uh oh!

drammock left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

aman-coder03 commented Mar 18, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants