DylComp module

class DylComp.Comparator(objects: list = None, level: int = 3, rand: bool = False, seed: int = None)

Bases: object

A class for comparing 2 values.

Controlled with the optimizaiton level and if you want random decisions or not Either provide objects in init or call .genLookup before you do any comparing Optimization levels: will not optimize, store result, do abc association, do recursive association Rand: defaults to False. If True will create data from random distributions with seed parameter

clearHistory()

Clears the history statistics of comparisons.

empiricROC() → dict

Generates and stores the empiric ROC if it needs to.

Returns the stored ROC curve.

genLookup(objects: list)

Generate the lookup table for each object provided.

genRand(n0: int, n1: int, sep: float, dist: str)

Generates the random data. If a seed has not previously been provided, it will be assigned here.

This new seeding may not work on Windows, so Windows users should assign the seed on their own.

genSeps() → list

Goes through the stored records and returns a list of the minimum separations.

If there is no minimum separation (the image has not been seen more than once), uses 2*(n0+n1) as a palceholder

getLatentScore(imgID: int) → float

gets the latent score of a given imgID or array of imgIDs.

If only one index is provided, also returns if the image is from the disease negative distribution.

kendalltau(predicted: list) → float

Returns the kendalltau statistic between the predicted image ID ordering and the true ordering of the image IDs with respect to latent score.

This method filters image IDs by what’s in predicted, so only the ids in predicted are used.

learn(arr: list, img: int = None, maxi: bool = False)

Learn the order of the array provided.

assuming the current optimization level allows it: if img is provided, learns the arr w.r.t. the img and if it is max or min. arr can also be a filename, in whichcase it will read the file to learn

max(arr, tryingAgain=False) → Tuple[int, int]

Gets the maximum of the array with respect to the latent scores.

tryingAgain should always be False unless a network comparator is used. Returns the undex of the maximum ID and the maximum ID.

min(arr) → Tuple[int, int]

Gets the minimum of the array with respect to the latent scores.

Returns the undex of the minimum ID and the minimum ID.

static optimize(objects: list, lookup: dict, res: bool, a, b) → int

Recursive optimization algorithm for adding a node to a fully connected graph.

Returns the number of optimizations it did.

record(vals: list)

Record that these values were seen.

This is automatically called by min and max.

resetPC()

Resets the pc statistics. Call this once per layer if you only want that layer’s PC.

updatePC(arr: list, guess, answer)

If the ids in arr are from different distibutions, adds 1 to the pc denominator.

If the guess was the answer, adds 1 to the pc numerator.

class DylComp.NetComparator(ip: str, port: int, recorder=None, objects: list = None, level: int = 3)

Bases: DylComp.Comparator

A class for doing comparisons over a network.

max(arr: list, tryingAgain=False) → Tuple[int, int]

Gets the maximum of the array with respect to the latent scores. tryingAgain is only used for if there was a hiccup in the network. Returns the undex of the maximum ID and the maximum ID.

min(arr: list) → Tuple[int, int]

Gets the minimum of the array with respect to the latent scores as the opposite of the maximum.

Returns the undex of the minimum ID and the minimum ID.