This is with the mistaken 1414 params, and not 1144.

(This seems right: https://github.com/tfiers/phd/blob/abb8db4/pkg/VoltoMapSim/src/params.jl ; but yes, this was the reason in the next notebook 1144 has same frs for inh and exc)

We'll be plottin F1max / max F1 / F_{1,\max}, F_1^*.

Why? Can't take recall, cause can make arbitrarily high. A [worry]: comparing apples and oranges: high R low P vs inverse, but same F1max. But (1) whatever; need two numbers if you want that info; and (2) is theoretic worry, in practice seems most classifiers are close to bisectrice.

South slavic

https://www.eurochicago.com/wp-content/uploads/2021/02/eurochicago20210130-052.jpg

src: https://github.com/tfiers/phd/tree/727fee4/pkg/VoltageToMap/src

this should be v_peak rather (35 mV, instead of –40 mV). But it works out in this case.

Codebase around finishing time of this notebook: https://github.com/tfiers/phd/tree/da6bc5bc1/pkg/VoltageToMap/src

(More commits around that time: https://github.com/tfiers/phd/commits/main?until=2022-06-02&since=2022-05-02&per-page=100 )

mistake: a minus in front is missing (or equivalently, replace the log with \( \log( \frac{τ_1}{τ_2}) \).

The maximum value on the line below it is correct however.

*similar enough

positive semi-definite

Newer and shorter version of this tutorial: https://timholy.github.io/SnoopCompile.jl/stable/tutorial/

https://tfiers.github.io/phd/nb/2022-09-09__Conntest_with_template_matching.html#:~:text=plotSTA(132%2C%201%2C%20s%2Cp)%3B%0AplotSTA(337%2C%201%2C%20s%2Cp)

This is data for one recorded/postsynaptic neuron (only five FPs)

note: these concrete neuron numbers are from a previous run with a different initialization, and do not correspond to the plot anymore.

In the current plot it's neuron 14 and 18

Also, found a bug where most hidden unit spikes seem to occur in pairs

Every dot is one inhibitory neuron (10 recorded, so 10 dots). The value of a dot is over all its inputs.

@Karim I saw this notebook didn't render on the website: I added a title in the first cell # .. and now it appears here. Also I saved the diagram as a separate image (in research/diagrams/) and referenced it (pasting doesn't work on the website it seems)

out of date, should be N_eval_trains

this plot shows SNR = 10. the results further down have SNR ∞, i.e. no noise

nope, k is in mS/mV, not mS. So use Wulfram's \(τ_m = RC\) rather

exp term becomes 0 for \(u ≤ θ_{rh}\), and ∞ when >. So downwards slope is the always added leak term.

Still puzzled about this. A neuron with 2 inputs has smaller STA's than one with 9 or 27 inputs. Difference is small though, and recording very noisy.

.

.

.

add comment: # Whether something ("a predicate") is true for each pair of neurons.

correction: 150 incoming neurons (300 total, p_connected 0.5)

Maybe because ROC-AUC is not an ideal measure: unfair for low # positives (low p_connected) vs low # negatives (high p_connected)

This should just be a the p-values in connection_test_summaries; no need to guess where the breaks are

Do not use this with tokenize()

https://bugs.python.org/issue23297#msg341028

Instead use the built-in open with "rb"

remove blank line (fuck black)

either return NotImplentedError or a comment ("# For subclasses to implement")

(also for _raised_to)

just nest under previous bullet (i.e.: "also if that data already has units ({examples})"

insert blank line

Insert:

    #
    #
    # -----------------
    # Unit composition

not consistent with the other dunders that just return nothing (None) when not implemented

This should be:

1 * unit == 1 * unit.data_unit * unit.conversion_factor

replace w/ 8 * mV

..yet

Give example (8 * mV or "8 mV")

just replace with

if not format_spec

(but add note that it's not None, but rather "")

replace with like:

`f"f-strings such as these: {array}"`

just array

move to after stmt (also in simple_unit.py)

idem

move to after stmt

oops, fix this typo

this is actually weird that it's the same

cause

\(3 * (6 / 2) = 3*3 = 9 ≠ \\ (3 * 6) / 2 = 18/2 = 9\)

oh wait it is I'm daft

== siemens

maybe call this data_conversion_factor ?

should have parameter to specify data_unit

(unprefixed unit needn't be the data_unit too; eg people may want to store data in mV)

better: numeric_equals

@norma

Like german "mundraub"

👀

This is seriously neat

Testing annotating Jupyter Book