adding a util function to calculate decaying reward-rate by hagikent · Pull Request #65 · AllenNeuralDynamics/aind-dynamic-foraging-basic-analysis

hagikent · 2025-05-31T23:27:21Z

A function to calculate decaying reward-rate time-series based on discrete reward events.
Exponential or hyperbolic kernel options are available.

tau: should be empirically optimised based on RT of mice etc
dt: if too small, even fftconvolve gets very slow. For now default ~50ms that corresponds to FIP effective sampling frequency, works with a reasonable computational load.

A function to calculate decaying reward-rate timeseries based on discrete reward events. Exponential or hyperbolic kernel options available. tau: should be empirically optimised based on RT of mice etc dt: if too small, even fftconvolve gets very slow. for now default ~50ms that corresponds to FIP effective sampling freqenncy.

rachelstephlee · 2025-06-02T17:22:18Z

+    ----------
+    event_times : (N,) array-like
+        1-D array of reward timestamps [s].
+    t_start, t_end : float, optional


would there be a reason to change t_start/t_end to a different time? is there value including this as an input?

For the current usage, not really, but if you have multiple sessions in a recording and want to partially calculate RewRate, this is handy. A good option to have as default=None.

keep this as it is-- i think we are going to consider multisession analysis in a different capsule. https://github.com/AllenNeuralDynamics/aind-dynamic-foraging-multisession-analysis

rachelstephlee · 2025-06-02T17:23:10Z

+
+    Parameters
+    ----------
+    event_times : (N,) array-like


should be renamed reward_time if this is specific to reward_rate-- if we just want to convolve this to any type of rate, we should rename this function instead.

good point. when/if aversive component is integrated, we can do the same excersize to calculate punishment rate, and I was picturing it.
Maybe reward_rate.py, reward_rate() -> event_rate?

i would make it event_rate instead. it should be general, so long that the tau, hyper_p are not tuned for reward, but for the trial length

according to your comment above, the latter seems correct.

can you provide documentation for the tau? or can you add a new issue to describe a function to map RT --> tau values? would be nice to add to this PR but we can keep this PR as it is also

I haven't worked on optimizing tau using RT. I will require a lot of dataset and should be separated from this as it's biology (not a part of util-prep).
will change the name.

hagikent · 2025-06-02T17:28:31Z

will lint soonish

rachelstephlee · 2025-06-02T17:32:59Z

+def reward_rate(event_times,
+                t_start=None,
+                t_end=None,
+                dt=1/20, # matching photometry frame rate


move these comments to the comment block below, so folks can see this when they run 'help' with the function

rachelstephlee · 2025-06-02T18:23:15Z

we should move this to metrics folder (or even metrics > trial_metrics.py file) since this is a trial metric and the file already calculates response rate (rolling, not convolved reward rate)

You can keep the inputs, you'll just need to pull reward events from df_trials (should be reward_time_in_session ; reward_outcome_time is usually the end of trial (regardless of whether or not the animal is rewarded)

See:

def compute_trial_metrics(nwb):
    """
    Computes trial by trial metrics

    response_rate,          fraction of trials with a response
    gocue_reward_rate,      fraction of trials with a reward
    response_reward_rate,   fraction of trials with a reward,
                            computed only on trials with a response
    choose_right_rate,      fraction of trials where chose right,
                            computed only on trials with a response
    intertrial_choice,      whether there was an intertrial lick event
    intertrial_choice_rate,   rolling fraction of go cues with intertrial licking
"""

minor filename (reward_rate->event_rate), location (into metrics)

rachelstephlee · 2025-06-02T19:52:07Z

discussing with @hanhou he agreed the files should be moved in metric, but forgaing_efficiency should be in a different PR with plenty of notice to folks.

linting using Black

rachelstephlee · 2025-06-02T21:44:58Z

+from scipy.signal import fftconvolve
+
+
+def event_rate(


one more- to match foraging_efficiency.py and trial_metrics, this function should be called compute_event_rate instead

rachelstephlee

looks good-- i ran it and it matches to what you've shown in our group chats.

i would fix the naming of the function and the final linting (add comment at the start of the file) and we can merge.

rachelstephlee

approved, thank you!

hagikent requested review from hanhou and rachelstephlee May 31, 2025 23:27

rachelstephlee reviewed Jun 2, 2025

View reviewed changes

file name, location change

d9cd2b0

minor filename (reward_rate->event_rate), location (into metrics)

hagikent added 3 commits June 2, 2025 13:38

lint

4dabaa6

linting using Black

minor re-locating foraging_efficiency.py

57f79de

lint

ea1d60b

rachelstephlee reviewed Jun 2, 2025

View reviewed changes

rachelstephlee requested changes Jun 2, 2025

View reviewed changes

lint, function name change, adding comments

32922f3

hagikent requested a review from rachelstephlee June 3, 2025 01:59

rachelstephlee approved these changes Jun 3, 2025

View reviewed changes

hagikent merged commit f28f8fd into main Jun 3, 2025
3 checks passed

hagikent deleted the kh-dev_reward-rate branch September 25, 2025 01:00

hagikent restored the kh-dev_reward-rate branch September 25, 2025 01:00

Uh oh!

Conversation

hagikent commented May 31, 2025

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!

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!

hagikent commented Jun 2, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rachelstephlee commented Jun 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rachelstephlee commented Jun 2, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rachelstephlee left a comment

Choose a reason for hiding this comment

Uh oh!

rachelstephlee left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rachelstephlee commented Jun 2, 2025 •

edited

Loading