Numerical series density - page 24

[Vyacheslav Kornev]

Well, again, why isn't the method I showed called v2 suitable? Which delta has the most clusters is the one to stop at.

[Vyacheslav Kornev]

Vyacheslav Kornev:
We have 50 cells and 11 dice with numbers

1, 3, 6, 8, 10, 11, 15, 16, 30, 40,50

V1. The densest clusters are: 10,11 и 15,16.

V2. Less dense are: 1,3 and 6,8,10,11 and 15,16

V3. Even less dense are 1,3,6,8,10,11 and 15,6

V4. Then 1,3,6,8,10,11,15,16.

The bottom line is this. We have picked up the delta. That is, we calculate v2 because in this variant there are the most clusters

Aggregation 1,3 takes 3 cells out of 50, i.e. 1.5 cells per cube.

Cluster 6,8,10,11 takes up 6 cells. And here 1.5 cells per cube. I won't go any further.

You didn't want to put 10 and 11 in a separate cluster.

See if delta 2 is the most common. And the centre of mass among the deltas here?

[Avals]

Until you define a clear density formula and a formula for comparing the two partitions into these clusters, you can break as many copies as you like. That is, there must be a numerical criterion for comparing the performance of the two algorithms.

[Aleksey Vyazmikin]

Vyacheslav Kornev:
Since you have understood that the bigger the delta the wider the cluster. Why do you say they are left-handed? Within a large cluster there are a lot of small ones

You are right - there are many small ones - perhaps there is a logical error in my algorithm - I need to think.

Vyacheslav Kornev:
Ah, up to what delta to count,

Well, heh, of all the deltas.

The most common.

Too much, though may be another estimation on other data.

Vyacheslav Kornev:

And generally by the method of finding the centre of mass. That is, count deltas among deltas.)

Um, how do you envisage it - state the algorithm.

[Aleksey Vyazmikin]

Vyacheslav Kornev:

See if delta 2 is the most common. And the centre of mass among the deltas here?

Further theoretical research is complicated due to lack of sufficient number of tests and evaluation of their results.

Without the code, I don't really want to kill time working in Excel right now - not productive.

Bottom line - the algorithm is to be tested on the data in order to test its effectiveness.

[Aleksey Vyazmikin]

Avals:
Until you define a clear density formula and a formula for comparing the two partitions into these clusters, you can break as many copies as you like. That is, there must be a numerical criterion for comparing the performance of the two algorithms.

Suggest your options for numerical estimation - I mentioned two options earlier.

[Aleksey Vyazmikin]

Changed the script code, now you can see the results directly on the chart - maximum density is highlighted in a separate colour, bar period can be selected by date or by bars from zero - by default.

[Aleksey Vyazmikin]

Added modified filter logic - seems to be less noise - changed by Variant parameter - 0 is old , and 1 is new (default).

[Aleksey Vyazmikin]

Added a filter based on the number of items in the group - 5 - can be changed. The more digits in the numeric row, the higher the value of the filter should be - it may be worth making a percentage of the maximum number of digits in the largest group.

Returned the previous default pre-filtering algorithm for the numeric row - 0.

[Aleksey Vyazmikin]

Corrected the calculation by shifting the data for the calculation.