Creating Dictionary of Dictionaries

Please Login or Register  to view this content.

Why would you want a dictionary of dictionaries, as opposed to a collection of dictionaries?

Would you think a collection of dictionaries is better?

You can access a collection by index; a dictionary only by key.

Originally Posted by kalak

There's not that much that I know of, although in general a dictionary of dictionaries shouldn't be too different from a dictionary of anything else. Also, the fact that you raised the topic implies you must have some idea of just why you did so.

For more information, there's a good summary of dictionaries by snb here http://www.snb-vba.eu/VBA_Dictionary_en.html , which also has something to say about indexes and keys.


It's unclear just what you mean by
"... identify similarities among the items in these baby dictionaries ..."
What kind of similarities?
Could you be more specific?
Example perhaps?

Thanks for the link. I actually went over about half of the content yesterday and checked the topics of the other half. It is a very comprehensive summary and very useful. Unfortunately, no direct info about dictionary of dictionaries. Probably I may refer the way they use to handle dictionary of arrays.

Sorry about the confusion caused by my description about my next step. It is a research project. What I need is to:

1. Store all data into dictionaries one a data set.

2. Compare between each pair of the dictionaries to identify identical items (and different items as well) using something like the MATCH function in Excel (suggestion of a better method?)

3. Split the items in each of the two dictionaries into two groups (identical ones and different ones between the two dictionaries). I will need to store the groups into new baby dictionaries.

4. Compare items in each with those in the remaining 99 dictionaries and to further split the groups.

5. All 100 data sets need to be analyzed in this way. Therefore, the number of combinations will be huge (any better idea than combination?).

6. The final groups should be undividable---indicating that items in each of these final groups co-exist in the nature.

Note: the Items in each data set are a part of a big data set. Items co-exist in the nature have less chance to be divided into different subsets.

Hope this is clear

You describe the steps of an algorithm. More generally, what are you trying to do?

It seems there should be a much easier way.

Some kind of DNA alignment thing?

What does the source data look like? What does the output look like?

Originally Posted by shg

What does the source data look like? What does the output look like?

Hi shg,

Sorry about replying you late. I didn't notice this 2nd page. The source data looks like these (only 6 subgroups from two rulers are shown):

Data.JPG

Where each R+3 digit number represents the number in the real order. ID is the "letters" on the ruler, but numbers are used instead here . Although the ID's are sorted based on their numeric values, they don't implicate anything about the real order of the "letters." G: subgroup; and "+": present.

I don't have a clear idea about the output format yet. It would be ideal if the output looks like these:

Result.JPG

You can see now, although the real order has not been recovered, the "letters" are grouped into segments with adjacent "letters" grouped together. Eventually, we should be able to deduce the order of the letters like this, in which the real order is recovered:

Result2.JPG

I tried to paste the tables in text in case you want to play with them. However the format can't be kept. So, they are all in images.

You can post a workbook.

Are those two examples side by side in each case, or two parts of a single example?

Is that two examples or one? If it's two, please get rid of one, and show the steps that transmute one set of inputs to one output.

Then I'll ask the logic applied at each step.

Please consider that only one of us has any clue of what you have or are trying to do.

Those are two SAMPLES in one EXAMPLE. Remember, we have 10 rulers (two are shown in the spread sheet) each was chopped at random points into 100 segments. The 100 segments are mixed first and then subdivided into 10 subgroups (6 (columns) are shown here.

[shg] On what basis are they assigned to subgroups? The + in the column is the subgroup assignment?Why are the column orders different in the two samples?

There are should be an average of 10 segments in each subgroup. Only one segment in each group is shown in the sheet. The steps to transmute the input information to output would include the follows:

1. Ordinarily, we may take a pair of "letters" (say 121 and 123) identified from the same subgroup (a column in the worksheet) and then ask if they are co-present among the subgroups from other nine rulers. If their co-present rate is significantly greater than the expected based on random combination, they are considered to be from the same segment, and otherwise, from different segments.

[shg] Where do you see "121" or "123"? You mean the ID column? Are the IDs the same on all rulers?

2. After exhausting all pairwise tests, letters of the same segment origin can be identified and sorted into segment-like form like what you see in the left two panels in the 2nd sheet, "Output & Final."

Note: The number of pairwise combinations is large. There should be method(s) similar to or better than the MATCH function that does not require pairwise test.

3. Run the same test for all subgroups from all 10 rulers so that letters in each group can be grouped based on their segment origins. At this point, you can only subdivide the letters based on their segment origins and don't know the order of the segments.

4. Work out the order of the overlapping "segments," for example

12 24 56 6 from one subgroup of ruler 1 and 56 6 897 25 from a subgroup of ruler 5 can be used to assemble the order of

12 24 56 6 897 254 based their overlapping letters of 56 and 6.

[shg] Where does 12 24 56 6 appear in a subgroup of ruler 1?

5. Extend the segment assembly until the entire order of the letters are worked out along the length of the entire ruler.

These steps are what I can imagine. There might be modifications as the programming goes.

[shg] I fear I am at sea. What's the genesis of the problem?

Those are two SAMPLES in one EXAMPLE. Remember, we have 10 rulers (two are shown in the spread sheet) each was chopped at random points into 100 segments. The 100 segments are mixed first and then subdivided into 10 subgroups (6 (columns) are shown here.)

[shg] On what basis are they assigned to subgroups? The + in the column is the subgroup assignment?Why are the column orders different in the two samples?


They are just randomly distributed into subgroups. As mentioned in the previous post, the reason for doing so is for the association test since closely located letters are tend to be distributed into the same subgroup. The + sign means "present". As show in the table on the left side, in the G1 column, 5, 10, 12, 45, 51, 54 and 982 are present while others are not. I shuffled the column order in the table on the right just want to indicate that the order of the column are irrelevant to the issue and does not provide any information for solving the problem.


There are should be an average of 10 segments in each subgroup. Only one segment in each group is shown in the sheet. The steps to transmute the input information to output would include the follows:

1. Ordinarily, we may take a pair of "letters" (say 121 and 123) identified from the same subgroup (a column in the worksheet) and then ask if they are co-present among the subgroups from other nine rulers. If their co-present rate is significantly greater than the expected based on random combination, they are considered to be from the same segment, and otherwise, from different segments.

[shg] Where do you see "121" or "123"? You mean the ID column? Are the IDs the same on all rulers?

Sorry about not taking the actual "letters" (IDs) from the tables. "121" and "123" were just examples. Yes, IDs are the same and in the same order on all rulers.

2. After exhausting all pairwise tests, letters of the same segment origin can be identified and sorted into segment-like form like what you see in the left two panels in the 2nd sheet, "Output & Final."

Note: The number of pairwise combinations is large. There should be method(s) similar to or better than the MATCH function that does not require pairwise test.

3. Run the same test for all subgroups from all 10 rulers so that letters in each group can be grouped based on their segment origins. At this point, you can only subdivide the letters based on their segment origins and don't know the order of the segments.

4. Work out the order of the overlapping "segments," for example

12 24 56 6 from one subgroup of ruler 1 and 56 6 897 25 from a subgroup of ruler 5 can be used to assemble the order of

12 24 56 6 897 254 based their overlapping letters of 56 and 6.

[shg] Where does 12 24 56 6 appear in a subgroup of ruler 1?

Sorry again, these are just examples, not taken from the actual subgroups in the table.

Let's work with the information from the tables in Sheet "Output & Final,"


In the left table,
G4 column has 218 7845 21 68 132 651 284 3 108

In the right table,
G5 column has 284 3 108 84 121

Sharing 284 3 108 by the two subgroups allows deduction of an order of

218 7845 21 68 132 651 284 3 108 84 121





5. Extend the segment assembly until the entire order of the letters are worked out along the length of the entire ruler.

These steps are what I can imagine. There might be modifications as the programming goes.

[shg] I fear I am at sea. What's the genesis of the problem?

It is a rather complex problem. It took me a while to figure out how to explain to you. The example that I give to you was a lot simplified from the actual data. However, it does present one of the core issues that I’m working on. If you read what I have described so far from the very beginning, you may start to get some sense about the issue. I'll be happy to answer more questions if you have

I reckon I feel that it's a complicated problem, and have no sense of the merit or value in trying to solve it, given a very abstract problem statement and no sense of what it all means.

Originally Posted by nilem

Hi Dodo123,
I just sort tables by the 1st column in a "Set2a" file and got exactly the same tables as in the "Set2b" file. And I see no reason to use a dictionary or a dictionary of dictionaries to get this result. Sorry.

Again, the 1st column is used to validate your results. It should be taken out and cannot be used to sort as it does not exist in the real data. You need to use logic association of the markers in the segments rather than using a pure algorithmic method to get the order, since this is not a magic cube.