Amazon Machine Learning works with CSV files of data. It doesn't appear to have any ability to work with relational data to represent one-to-many relationships.
How should I transform a relational dataset so that it can be used for machine learning?
Would it be best to denormalize the dataset or am I thinking about this the wrong way?
Your best bet would be to denormalize the dataset, so that each input observation has all the attributes (columns) needed to make a prediction. If you can provide a few example data rows, even using made-up data values, I'd be happy to help more.
Related
I'm new to PowerBI, and am working on a large database. I am attempting to prepare the data in the PowerQuery Editor.
I would like to code as many steps as possible, as analysing each column manually is extremely time consuming.
My coding goals (in order of priority):
For each query I would like to get their column quality.
Ideally, I would like to export the header names with the column quality, so that I can determine which are relevant. Furthermore, I can also use the column names to determine which column relationships might be relevant. The database is huge, so simply just importing all the data and trying to work with it from their is not feasible, in fact PowerBI comes up with the error that I don't have enough free memory.
I have VBA and some SQL experience.
I know I have a lot to learn w.r.t. PowerBI, and I am working on it, but need some guidance and direction, also on what is possible/feasible.
Any contructive hints, advice, or feedback would be appreciated - thank you!
Use Table.Profile() on each table and load to the data model.
https://learn.microsoft.com/en-us/powerquery-m/table-profile
I wanted to know what would be the best approach for creating the dim tables. Can I maintain it as a single table with all fields and use them as required or create separate dim tables and use them individually.
Can someone please help me out here
PS: I'm a beginner here.
Creating 1 table per dimension is the best practice. In data warehouse concept, you will get 4 types of schema as below-
Start Schema
Snowflakes Schema
Galaxy Schema
Combined Schema
People select any of the above based on their Data type/nature, requirement and other parameter. But in all case, there are single table per dimension. This is easy to maintain and give better performance.
Well I recently got into this area of Redshift, trying to optimize disk usage and performance of my database, and having read lots of information on AWS about the topic, I still have some doubts.
First of all, to my database structure. Per schema, I have 3 master tables, with 3 different IDs, these are now DISTSTLYE ALL tables, being small in size.
Each master table has different amounts of IDs,
the date table --> largest one (#1 most joined)
the store table --> medium one (#3 most joined)
the item table --> smallest one (#2 most joined)
Then I have my core table, which has needed combinations of these IDs to display additional information about them. Anyway, this table should be a DISTSTYLE KEY type, based on my knowledge. Well, which of the 3 IDs should I select to be my DIST KEY?
Whats the criteria for this decision? I understand that for joins I need to look at the Sort Key, well that has been understood and defined to the ID_date, because its the most joined table. So now, what about the distribution per node of this table?
I'm sorry if I'm rambling, I dont want to leave any information out. If I have, feel free to ask! Thanks for taking the time to read!
You'll find the best advice on Amazon Redshift best practices for designing tables. It goes into quite a bit of detail.
However, my rule of thumb is:
The DISTKEY should be the column most used in JOINs between tables
The SORTKEY should be the column most used in WHERE statements
Use DISTSTYLE ALL for small lookup tables
I have a table on Athena partitioned by day (huge table, TB of data). There's no day column on the table, at least not explicitly. I would expect that a query like the following:
select max(day) from my_table
would scan virtually no data. However, Athena reports that several hundreds of GB are scanned. Any idea why?
===== EDIT 2021-01-14 ===
I've recently bumped on this issue again. It turns out that when the underlying data is parquet then operations on partitions don't consume data. For other data formats that I've tried (including ORC) there is an associated data cost. It doesn't make any sense to me.
I don't know the answer for a fact but I guesstimate:
Athena just does not have the optimization of looking at the partition names only, when only they are queried. This is clear from its behaviour. So it scans everything.
Parquet has min/max for every column whereas ORC does it only if an index is present, AFAIU. Thus for Parquet Athena's query optimizer directs it to look directly at these rollup values, i.e., no scan is performed. It's different for ORC.
I know is a little late to answer this question for you Nicolas but it is important to keep here also some possible solutions.
Unfortunately, this is the way Athena works, Athena will read all data as a tableScan just to list the partitions values.
A possible workaround that works perfectly here is using the metadata of the partition instead of the data information, for example:
Instead of using this syntax:
select max(day) from my_table
Try to use this syntax:
SELECT day FROM my_schema."my_table$partitions" ORDER BY day DESC LIMIT 1
This second statement will read just metadata information and returns the same data you need.
It does not depend on the format but on the compression algorithm used. Snappy for ORC mostly & GZIP for parquet. This is what makes the difference
I am looking for solutions or ideas how to speed the processing of large data sets in sas.
What would you recommend?
What is better data step or proc sql procedure?
Speeding up your data processing depends on where your data is saved.
Your data can be either in:
SAS Table,
Database Table (Miscrosfot SQL, Oracle, DB2, MYSQL, ..
etc.)
Use SAS Data Step when:
You are querying/processing SAS tables,
You want to do iterative
processing (ex. retaining values or using arrays).
Use Proc SQL when:
You are querying a large Database table,
You can do a SQL "Pass Through" where you send SQL code to be
executed on the DB server and only the output is sent to SAS (instead
of bringing the entire tables through the network to SAS and then filter it),
You want to query SAS Tables but prefer SQL joins to data step merges.
Another topic you should consider is efficiency programming; where you are optimising your query and look-ups.
I find Proc SQL to be better for my use cases. We may need some more specifics on the size and variety of data your trying to join/export etc.
Give us some info on that and we can try to help.
Tips:
Limit the fields your pulling over
Subset data
Anecdotally from my experience Proc SQL seems faster.
Here are two tips on speeding up queries with Proc SQL:
In general, you want to rule out as much data as possible when querying. If you are usingProc SQL, the order of the restrictions in the where clause matters. Put the most restrictive parts first.
For example, if I'm querying a database for teachers with the last name "JONES", that were hired after Jan 2005, I would structure my where clause like this: where last_name = 'JONES' and hire_date > 200501 I would do this because last name is likely to exclude more records than the hire date restriction.
When possible, don't use Select * instead, list out the specific columns that you need. Remember, even if you are doing a calculation with a column, you don't have to include that column in your select statement.
Here is a very useful resource for understanding how to use proc sql efficiently. I recommend reading it in it's entirety if you do a lot of work with large data sets in SAS.
http://www2.sas.com/proceedings/sugi29/127-29.pdf