FeSi Profiles
FeNi Profiles
TeNi Profiles
TeSi Profiles
NeTi Profiles
SeTi Profiles
SeFi Profiles
NeFi Profiles
TiNe Profiles
TiSe Profiles
FiSe Profiles
FiNe Profiles
SiFe Profiles
NiFe Profiles
NiTe Profiles
SiTe Profiles
Raw data analysis, and avoiding self confirmation bias

Home 2023 Forums General Psychology Raw data analysis, and avoiding self confirmation bias

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
    Posts
  • #25750
    Starshade
    Participant
    • Type: NeFi
    • Development: ll-l
    • Attitude: Seelie

    Some years ago I found an personal puzzle which, took me 2-3 years to figure out: how do an Humanities/Sosial Antrophology theory focused Cultural History student, with zero STEM, read DNA data well enough to find Saami heritage using Myheritage?
    The answer: I needed help of STEM. And secondary, learn to work with the genetics data. And I started learning how bad some publications handle their data sets, and how hobbyists often can mismanage theirs.
    Ok, In order to make an system, the DNA people seem to use several data methods, of which I found a few interesting.
    Self Confirmation bias, is using your own type data used to make the system, to test your own system. This is a Phd level typology issue, which, can affect quite advanced research; it can affect any data driven type system, therefore my post on this.
    What I have learned, is the STEM approach, is this:

    • Select a set of core data for making your system, and de-select a few good analytical data, of which will be set aside to test your finished product.
    • Make the system.
    • Test the new system, with your safe and sure data, for which you know should work as espected.

    The biggest differences is CT 1.0 uses a set of 110 datapoints where DNA companies goes in the range of 3 thousand data points. But thought I should share some thoughts on this; the flaw, is, with this idea, you need an external souce of data, to use this methodology.

    #25752
    unknown
    Participant
    • Type: Unknown
    • Development:
    • Attitude: Unknown

    using your own type data used to make the system, to test your own system
    you need an external souce of data, to use this methodology.


    What data we need here to fix the issue?
    I am curious how the process of selecting (datapoints) signals looked like and what reasoning stand behind it? How emotional body and face expressions turn into indicator for cognitive funtcions which from science point of view are not confirmed and described? I know signals have explanations but I can't see direct connections. Maybe there is explanation I didn't found yet. It's looks like we have an assumtion "this is it and we will work with this".

    #25753
    Juan E. Sandoval
    Keymaster
    • Type: TiNe
    • Development: l--l
    • Attitude: Adaptive

    These are good questions!
    They're partly addressed here: https://cognitivetypology.com/index.php?title=Motus_Project
    Coming Up with a Set of Signals
    In brief, the question of whether or not a given selection of signals (i.e. the CT vultology code or someone else's) was void of bias is something I don't think human eyes on their own can measure very well -- because of how multivariable it is.
    I believe that we would need to create a software program that could be fed thousands of videos, and do statistical analysis on those videos in order to naturally and atheoretically derive the most robust repetitive signals, as well as which signals are most statistically linked to each other. I tried to do this myself -- it was the very first thing I tried to do in a pre-CT era, but the endeavor is just too labor intensive. So I think computers need to decide that.
    Ideally, if a program like this was made, we could see whether CT's codex is reflected in those results. So you could check CT against that metric. But this doesn't yet exist, as far as I'm aware, although computer vision is getting awesome lately, so hopefully it will happen soon.
    For now
    Given that we don't yet have a purely objective way of knowing whether a codex is a natural representation of human expressivity, we can look instead at the results. For example, do CT's chosen categories lead to meaningful data parsing when applied across wide populations? If the answer is yes, then we have reason to believe these divisions are not random, or that the CT codex is aligning itself to a natural phenomenon.
    I'm working on more and more ways to support this with evidence. You can look at this page for some information: https://cognitivetypology.com/index.php?title=Occupations
    What you'll see in that page is that when these signals are applied to a population of 400+, they produce statistically significant career/occupation clusterings. This is just one of the proofs-of-concepts. Survey data also seemsĀ  to support the non-randomness of the signal sets. That's a pilot study that's current about halfway done, and I hope to publish that one this month. It shows that Pe-Pi-Je-Pi scores on that survey correspond to people's vultological typing.
    So for the time being I'm taking a pragmatic approach to the efficacy of the vultology code and proposing that there is merit to it, because it works.
    It can be tested, and its results align with predictions more often than chance.

    #25760
    Starshade
    Participant
    • Type: NeFi
    • Development: ll-l
    • Attitude: Seelie

    That, sounds good.
    My thoughts on data AI video analysis:

    • It might need to have an reduced data set, but broad enough. Make something resembling the data from a Aircraft's black box. This is analysed by a learning data program.
    • This, I think, is achieveable if one makes markers for the "learner" program to follow, to get the set.

    Last part, I think, is easiest done by actually painting markers on the face. Could be a few dots, or full Blue/Black with white dots and stripes (no joking, I'm serious, for data learning).
    Base data could then be used for making face tracking software for normal faces.

Viewing 4 posts - 1 through 4 (of 4 total)
  • You must be logged in to reply to this topic.
bookmarkglobeenvelopeshopping-bag