Thread: Advice for using integer arrays?
I'm fairly new to Postgres, and have a design issue for which an array of integers might be a good solution. But I'd like to hear from the experts before proceeding down this path. Essentially, I'm trying to model the relationship between a group of biological samples and their genes. Each sample (there are ~10K of them now, with more coming) will have about 30,000 genes. Conversely, a particular gene may be present in almost all samples. So I've created the following table to handle the many-to-many relationship: create table sample_gene (id serial, sample_id int, gene_id int); which looks like this when populated: sample_id | gene_id --------------------------------------- 1 | 1 1 | 2 ... 1 | 30475 2 | 1 2 | 2 ... 2 | 29973 3 | 1 etc. The table now contains hundreds of millions of rows (with many, many more to come). Join performance between samples and genes is quite slow, even with indexes on sample_id and gene_id. So it occurred to me: why not eliminate all the duplicate sample_id values by storing the gene_id's in an array, like so: create table sample_gene_array (id serial, sample_id int, gene_id int [] ); So now the table data looks like this: sample_id | gene_id [] --------------------------------------- 1 | [1:30475] 2 | [1:29973] etc. The new table is significantly smaller, and performance (using ANY[] ) is quite good. Nevertheless, I'm uneasy. I come from a Sybase ASE background, and so have no experience with arrays as datatypes. Is it okay to store 30K+ gene values in an array in the linking table (or maybe even in the sample table itself, thus eliminating the linking table)? Should I unnest the gene_id's first, before using them to join to the gene table? TIA for any guidance you can provide. Again, I'm a Postgres neophyte - but I'm in awe of the power and flexibility of this database, and wish that I'd started using it sooner. ------ Michael Heaney JCVI
On Jan 6, 2015 3:12 PM, "Michael Heaney" <mheaney@jcvi.org> wrote:
>
> I'm fairly new to Postgres, and have a design issue for which an array of integers might be a good solution. But I'd like to hear from the experts before proceeding down this path.
>
> Essentially, I'm trying to model the relationship between a group of biological samples and their genes. Each sample (there are ~10K of them now, with more coming) will have about 30,000 genes. Conversely, a particular gene may be present in almost all samples.
>
> So I've created the following table to handle the many-to-many relationship:
>
> create table sample_gene (id serial, sample_id int, gene_id int);
>
> which looks like this when populated:
>
> sample_id | gene_id
> ---------------------------------------
> 1 | 1
> 1 | 2
> ...
> 1 | 30475
> 2 | 1
> 2 | 2
> ...
> 2 | 29973
> 3 | 1
> etc.
>
> The table now contains hundreds of millions of rows (with many, many more to come). Join performance between samples and genes is quite slow, even with indexes on sample_id and gene_id.
>
> So it occurred to me: why not eliminate all the duplicate sample_id values by storing the gene_id's in an array, like so:
>
> create table sample_gene_array (id serial, sample_id int, gene_id int [] );
>
> So now the table data looks like this:
>
> sample_id | gene_id []
> ---------------------------------------
> 1 | [1:30475]
> 2 | [1:29973]
> etc.
>
> The new table is significantly smaller, and performance (using ANY[] ) is quite good. Nevertheless, I'm uneasy. I come from a Sybase ASE background, and so have no experience with arrays as datatypes. Is it okay to store 30K+ gene values in an array in the linking table (or maybe even in the sample table itself, thus eliminating the linking table)? Should I unnest the gene_id's first, before using them to join to the gene table?
>
> TIA for any guidance you can provide. Again, I'm a Postgres neophyte - but I'm in awe of the power and flexibility of this database, and wish that I'd started using it sooner.
>
> ------
> Michael Heaney
> JCVI
>
>
>
>
>
> --
> Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-general
Please provide sample queries so we can understand how you query the data.
On 01/06/2015 10:09 AM, Michael Heaney wrote:
I'm fairly new to Postgres, and have a design issue for which an array of integers might be a good solution. But I'd like to hear from the experts before proceeding down this path.Think genotype calls (sample, marker, call) and this just explodes in terms of number of rows but I'm trying to not let that bother me. Partion by chromosome, or chromosome arm (even if partition is just separate tables) and the magnitude becomes manageable. At least as manageable as multi-gigabyte gvcf files per sample from GATK.
Essentially, I'm trying to model the relationship between a group of biological samples and their genes. Each sample (there are ~10K of them now, with more coming) will have about 30,000 genes. Conversely, a particular gene may be present in almost all samples.
So I've created the following table to handle the many-to-many relationship:
create table sample_gene (id serial, sample_id int, gene_id int);
which looks like this when populated:
sample_id | gene_id
---------------------------------------
1 | 1
1 | 2
...
1 | 30475
2 | 1
2 | 2
...
2 | 29973
3 | 1
etc.
The table now contains hundreds of millions of rows (with many, many more to come). Join performance between samples and genes is quite slow, even with indexes on sample_id and gene_id.
So it occurred to me: why not eliminate all the duplicate sample_id values by storing the gene_id's in an array, like so:
create table sample_gene_array (id serial, sample_id int, gene_id int [] );
So now the table data looks like this:
sample_id | gene_id []
---------------------------------------
1 | [1:30475]
2 | [1:29973]
etc.
The new table is significantly smaller, and performance (using ANY[] ) is quite good. Nevertheless, I'm uneasy. I come from a Sybase ASE background, and so have no experience with arrays as datatypes. Is it okay to store 30K+ gene values in an array in the linking table (or maybe even in the sample table itself, thus eliminating the linking table)? Should I unnest the gene_id's first, before using them to join to the gene table?
TIA for any guidance you can provide. Again, I'm a Postgres neophyte - but I'm in awe of the power and flexibility of this database, and wish that I'd started using it sooner.
------
Michael Heaney
JCVI
But I'm a little confused: all samples for a given species will have the same genes (roughly). Are you storing gene variants (sequence or otherwise) per sample?
On Tue, Jan 6, 2015 at 9:09 AM, Michael Heaney <mheaney@jcvi.org> wrote:
I'm fairly new to Postgres, and have a design issue for which an array of integers might be a good solution. But I'd like to hear from the experts before proceeding down this path.
Essentially, I'm trying to model the relationship between a group of biological samples and their genes. Each sample (there are ~10K of them now, with more coming) will have about 30,000 genes. Conversely, a particular gene may be present in almost all samples.
So I've created the following table to handle the many-to-many relationship:
create table sample_gene (id serial, sample_id int, gene_id int);
What is the value of having this table at all? It doesn't seem to contain anything informative, like an allele identifier, a resequence, or a copy number variation. If you are just trying to record the fact that a gene was present in that sample, perhaps it would be better to instead record the genes have been deleted, rather than the ones that have not been deleted? That would probably be a much smaller list.
create table sample_gene_array (id serial, sample_id int, gene_id int [] );
So now the table data looks like this:
sample_id | gene_id []
---------------------------------------
1 | [1:30475]
2 | [1:29973]
etc.
I'm not familiar with the square bracket and colon as a syntax for expressing int arrays. Are you taking liberties with the psql output, or using a different client program? Does that represent the range from 1 to 30475, or the two values 1 and 30475?
Cheers,
Jeff
On 1/6/2015 2:19 PM, Jeff Janes wrote:
On Tue, Jan 6, 2015 at 9:09 AM, Michael Heaney <mheaney@jcvi.org> wrote:I'm fairly new to Postgres, and have a design issue for which an array of integers might be a good solution. But I'd like to hear from the experts before proceeding down this path.
Essentially, I'm trying to model the relationship between a group of biological samples and their genes. Each sample (there are ~10K of them now, with more coming) will have about 30,000 genes. Conversely, a particular gene may be present in almost all samples.
So I've created the following table to handle the many-to-many relationship:
create table sample_gene (id serial, sample_id int, gene_id int);What is the value of having this table at all? It doesn't seem to contain anything informative, like an allele identifier, a resequence, or a copy number variation. If you are just trying to record the fact that a gene was present in that sample, perhaps it would be better to instead record the genes have been deleted, rather than the ones that have not been deleted? That would probably be a much smaller list.
I suppose there could be a gene table which would contain data about each gene_id.
But I'm an IT guy, not a biologist, and my sample_gene table doesn't actually
exist. I'm more concerned with how to deal with many-to-many relationships
when each parent could have tens of thousands of children. Collapsing all
the children into an array for each parent looked intriguing - but maybe it's
not a good idea. I just don't know, so I thought I'd ask you guys.
create table sample_gene_array (id serial, sample_id int, gene_id int [] );
So now the table data looks like this:
sample_id | gene_id []
---------------------------------------
1 | [1:30475]
2 | [1:29973]
etc.I'm not familiar with the square bracket and colon as a syntax for expressing int arrays. Are you taking liberties with the psql output, or using a different client program? Does that represent the range from 1 to 30475, or the two values 1 and 30475?
Yes, it's shorthand for the full range of values from 1 to 30475. Wasn't sure how
to represent it, and almost went with (1,2,3...30475). Apologies...
Michael Heaney
JCVI
Hi Michael,
I can't comment on the domain-specific stuff, but I recently used numeric arrays for a project and it worked well. In my case we had one million simulation results (floats) per scenario, so rather than reading one million separate rows to compute a histogram, we stored everything in one row per scenario. Ideally one million floats is 8 megabytes, which is big but still shouldn't require more than 100ms to read from disk and feed into a simple computation. Here are some functions I wrote to make it easier & faster to compute stats from numeric arrays:
https://github.com/pjungwir/aggs_for_arrays/
One caveat is that you will lose a lot of benefits by leaving the relational model, e.g. being able to filter your integers by other criteria. In our case all we had were floats, but imagine if you had metadata attached to each one like the time the simulation was run, who ran it, etc. Then you'd want to stick with something richer than just an array of numbers.I can't comment on the domain-specific stuff, but I recently used numeric arrays for a project and it worked well. In my case we had one million simulation results (floats) per scenario, so rather than reading one million separate rows to compute a histogram, we stored everything in one row per scenario. Ideally one million floats is 8 megabytes, which is big but still shouldn't require more than 100ms to read from disk and feed into a simple computation. Here are some functions I wrote to make it easier & faster to compute stats from numeric arrays:
https://github.com/pjungwir/aggs_for_arrays/
But in that case maybe parallel arrays is acceptable. It would be a bit like a column-store inside of Postgres. :-) I've been meaning to add a function to that Github repo to filter an array given a same-size array of booleans, so you can do filtering like in R or Pandas, but I haven't found a nice way in Postgres to express e.g. `filter_array(simulation_results, simulation_run_times > now() - interval '1 day')`.
Good luck!
On Tue, Jan 6, 2015 at 1:18 PM, Michael Heaney <mheaney@jcvi.org> wrote:
On 1/6/2015 2:19 PM, Jeff Janes wrote:On Tue, Jan 6, 2015 at 9:09 AM, Michael Heaney <mheaney@jcvi.org> wrote:I'm fairly new to Postgres, and have a design issue for which an array of integers might be a good solution. But I'd like to hear from the experts before proceeding down this path.
Essentially, I'm trying to model the relationship between a group of biological samples and their genes. Each sample (there are ~10K of them now, with more coming) will have about 30,000 genes. Conversely, a particular gene may be present in almost all samples.
So I've created the following table to handle the many-to-many relationship:
create table sample_gene (id serial, sample_id int, gene_id int);What is the value of having this table at all? It doesn't seem to contain anything informative, like an allele identifier, a resequence, or a copy number variation. If you are just trying to record the fact that a gene was present in that sample, perhaps it would be better to instead record the genes have been deleted, rather than the ones that have not been deleted? That would probably be a much smaller list.
I suppose there could be a gene table which would contain data about each gene_id.
But I'm an IT guy, not a biologist, and my sample_gene table doesn't actually
exist. I'm more concerned with how to deal with many-to-many relationships
when each parent could have tens of thousands of children. Collapsing all
the children into an array for each parent looked intriguing - but maybe it's
not a good idea. I just don't know, so I thought I'd ask you guys.create table sample_gene_array (id serial, sample_id int, gene_id int [] );
So now the table data looks like this:
sample_id | gene_id []
---------------------------------------
1 | [1:30475]
2 | [1:29973]
etc.I'm not familiar with the square bracket and colon as a syntax for expressing int arrays. Are you taking liberties with the psql output, or using a different client program? Does that represent the range from 1 to 30475, or the two values 1 and 30475?
Yes, it's shorthand for the full range of values from 1 to 30475. Wasn't sure how
to represent it, and almost went with (1,2,3...30475). Apologies...
Michael Heaney
JCVI
--
_________________________________
Pulchritudo splendor veritatis.
Pulchritudo splendor veritatis.
On 01/06/2015 01:18 PM, Michael Heaney wrote: > On 1/6/2015 2:19 PM, Jeff Janes wrote: >> On Tue, Jan 6, 2015 at 9:09 AM, Michael Heaney <mheaney@jcvi.org >> <mailto:mheaney@jcvi.org>> wrote: >> >> I'm fairly new to Postgres, and have a design issue for which an >> array of integers might be a good solution. But I'd like to hear >> from the experts before proceeding down this path. >> >> Essentially, I'm trying to model the relationship between a group >> of biological samples and their genes. Each sample (there are ~10K >> of them now, with more coming) will have about 30,000 genes. >> Conversely, a particular gene may be present in almost all samples. >> >> So I've created the following table to handle the many-to-many >> relationship: >> >> create table sample_gene (id serial, sample_id int, gene_id int); >> >> >> What is the value of having this table at all? It doesn't seem to >> contain anything informative, like an allele identifier, a resequence, >> or a copy number variation. If you are just trying to record the fact >> that a gene was present in that sample, perhaps it would be better to >> instead record the genes have been deleted, rather than the ones that >> have not been deleted? That would probably be a much smaller list. >> > > I suppose there could be a gene table which would contain data about > each gene_id. > But I'm an IT guy, not a biologist, and my sample_gene table doesn't > actually > exist. Alright, now I am confused. In your original post you say sample_gene does exist and you joined it against genes(I assume a gene table) which is when you encountered slow performance. Did I miss something:)? Per a previous suggestion, it might be good to show the queries you are using or plan to use. The issue may be in the query not the layout. Also running the query with EXPLAIN ANALYZE would be helpful: http://www.postgresql.org/docs/9.3/interactive/sql-explain.html I'm more concerned with how to deal with many-to-many relationships > when each parent could have tens of thousands of children. Collapsing all > the children into an array for each parent looked intriguing - but maybe > it's > not a good idea. I just don't know, so I thought I'd ask you guys. > > > >> create table sample_gene_array (id serial, sample_id int, gene_id >> int [] ); >> >> So now the table data looks like this: >> >> sample_id | gene_id [] >> --------------------------------------- >> 1 | [1:30475] >> 2 | [1:29973] >> etc. >> >> >> I'm not familiar with the square bracket and colon as a syntax for >> expressing int arrays. Are you taking liberties with the psql output, >> or using a different client program? Does that represent the range >> from 1 to 30475, or the two values 1 and 30475? > > Yes, it's shorthand for the full range of values from 1 to 30475. Wasn't > sure how > to represent it, Use Python? Looks like a 1 based slice notation. and almost went with (1,2,3...30475). Apologies... > > > Michael Heaney > JCVI > > > -- Adrian Klaver adrian.klaver@aklaver.com
On Tue, Jan 06, 2015 at 12:09:56PM -0500, Michael Heaney wrote: > I'm fairly new to Postgres, and have a design issue for which an > array of integers might be a good solution. But I'd like to hear > from the experts before proceeding down this path. The biggest consideration is if you are ever intending to use the values in a join condition. Arrays a efficient space-wise and you also have good indexing strategies with GIN indexes. You will need to reframe your queries in terms of ([x] subset-of field) but that's relatively straightforward. What doesn't work or is fiddely: - foreign keys - selecting part of the list - reordering or otherwise manipulating the list. basically, if conceptually the list is a single object which you're really only going to want to access as a whole, but still want good indexing, then arrays are for you. BTW, looking at your example, you might be more interested in ranges, see for example: http://www.postgresql.org/docs/9.2/static/rangetypes.html Conceptually they are a bit different and there isn't support for multi-ranges AFAIK but they might be more appropriate. Hope this helps, -- Martijn van Oosterhout <kleptog@svana.org> http://svana.org/kleptog/ > He who writes carelessly confesses thereby at the very outset that he does > not attach much importance to his own thoughts. -- Arthur Schopenhauer
Attachment
> BTW, looking at your example, you might be more interested in ranges, > see for example: > http://www.postgresql.org/docs/9.2/static/rangetypes.html > > Conceptually they are a bit different and there isn't support for > multi-ranges AFAIK You could have an array of ranges