Posts

Recently, I encountered a python function that I didn’t want to modify, but whose output I wanted cached. Providing typehints for this seemed a bit complicated, so I’m posting this gist as a reminder. The main trick involves extensions from PEP 612. In words, the decorator will take some function f that takes parameters P and produces a str. I want to be able to call f and additionally provide a filename that will point to a file in which the string will be written.

How to store a NifTi as a TFRecord Patrick Sadil 2022-04-23 A recent project required sending brain images to TensorFlow. Unfortunately, the data exceeded memory and so during training would need to be read from the disk. Poking around the TF documentation, it seems like a recommended way to do this is to store the images as a TFRecord. The steps for doing that are collected in this gist1. There are five main steps

knitr::opts_chunk$set(echo = TRUE) set.seed(1234) library(dplyr) library(ggplot2) library(tidyr) I’m working on a project involving serial dependence. The project involves disentangling a dependence on the previous orientation from a dependence on the previous response. Unfortunately, there is a common way for a dependence on the previous response to be spurious, due to the oblique effect1. The first reference I’ve seen for this is a master’s thesis by Fritsche (2016). I didn’t follow that explanation, and so I’m using this post to explain how the oblique effect causes a spurious dependence on the previous response.

A friend recently moved away from Western Mass, so I wanted to send them a gift to help remind them of the area. I also happened to want to learn a bit more about Generative Adversarial Networks (GANs). Goals Briefly overview GANs Link to fun code for acquiring data from Google Street View Share a neat set of photographs Note that this isn’t an ‘intro to GANs’.

Many experiments require counterbalancing sequences of trials. For example, I’m currently running an experiment on serial dependence1. In my experiment, participants report the orientation of a grating2 stimulus on each trial. The serial dependence effect is how their responses on one trial depend on either the orientation of the previous trial or their response on that trial. To tease apart the effects of prior stimuli from prior responses, I’m manipulating the visual contrast of the gratings ( Michelson contrast ).

Visual perception research has produced many illusions. Stare at a waterfall for a minute, then look away and the whole world appears in motion, traveling upwards (Addams 1834). Given proper lighting, black paper can appear white, but placing a white piece of paper nearby colors the first dark gray (Gelb 1929; as cited by Cataliotti and Gilchrist 1995). Inspecting two sets of black lines – horizontal lines that obscure a solid red field like a picket fence, along with vertical lines that obscure a solid green field – causes the black lines alone to induce a perception of color, an illusory shading that can last for days (McCollough 1965).

Science produces models that provide parsimonious descriptions of the world. In cognitive psychology, models regularly compete to explain a few phenomena. But models can survive experiment after experiment, both because of the difficulty of capturing participant’s nuanced behavior, and because models often make highly overlapping predictions. In these cases, a model succeeds through its relative parsimony. In cognitive psychology, measures of information criteria, specifically Akaike’s and the Bayesian information criteria (Schwarz 1978; Akaike 1973), determine a winning model.

In last week’s post, I discussed how some experiments in cognitive psychology require researchers to pick a differently intense stimulus for each participant. In particular, I discussed a procedure for picking an intensity that elicits positive responses on approximately half of trials, the staircase procedure. In the staircase procedure, the researcher increases the intensity after every positive response and decreases the intensity after every negative response. If the participant completes another set of trials in which the intensity is fixed to the average of the intensities that were used during the staircase, the participant will provide positive responses approximately half of the time.

Performing any experiment on cognition requires deciding which stimuli to use. Some experiments require participants to make many errors, requiring the stimuli to be challenging. In other experiments, participants must respond accurately, requiring stimuli that are easy but not so easy that participants lose attention. Moreover, participants behave idiosyncratically, so to avoid wasting either the researchers’ or participants’ time the stimuli ought to be tailored to each participant. To decide which stimuli to use, researchers can rely on a psychometric function (Figure 1).

Functional magnetic resonance imaging records brain activity with spatially distinct voxels, but this segmentation will be misaligned with a brain’s meaningful boundaries. The segmentation results in some voxels recording activity from different types of tissue – types that are both neural an non-neural – but even voxels that exclusively sample gray matter can span functionally distinct cortex. For example, a 3T scanner allows voxels in the range of 1.