CASTable vs. DataFrame vs. SASDataFrame¶
CASTable objects and DataFrame object (either pandas.DataFrame or SASDataFrame) act very similar in many ways, but they are extremely different constructs. CASTable objects do not contain actual data. They are simply a client-side view of the data in a CAS table on a CAS server. DataFrames and SASDataFrames contain data in-memory on the client machine.
Since these objects work very much the same way, it can be a little confusing when you start to work with them. The rule to remember is that if the type of the object contains “DataFrame” (i.e., DataFrame or SASDataFrame), the data is local. If the type of the object contains “Table” or “Column” (i.e., CASTable or CASColumn), it is a client-side view of the data in a CAS table on the server.
Even though they are very different architectures, CASTable objects support much of the pandas.DataFrame API. However, since CAS tables can contain enormous amounts of data that wouldn’t fit into the memory of a single machine, there are some differences in the way the APIs work. The basic rules to remember about CASTable data access are as follows.
- If a method returns observation-level data, it will be returned as a new CASTable object.
- If a method returns summarized data, it will return data as a SASDataFrame.
In other words, if the method is going to return a new data set with potentially huge amounts of data, you will get a new CASTable object that is a view of that data. If the method is going to compute a result of some analysis that is a summary of the data, you will get a local copy of that result in a SASDataFrame. Most actions allow you to specify a casout= parameter that allows you to send the summarized data to a table in the server as well.
The methods and properties of CASTable objects that return a new CASTable object are loc, iloc, ix, query, sort_values, and __getitem__ (i.e., tbl[...]). The remaining methods will return results. A sampling of a number of methods common to both CASTable and DataFrame are shown below with the result type of that method.
In the table above, the CASColumn is a subclass of CASTable that only references a single column of the CAS table.
The last entry in the table above is a call to a CAS action called summary. All CAS actions return a CASResults object (which is a subclass of Python’s ordered dictionary). DataFrame’s can not call CAS actions, although you can upload a DataFrame to a CAS table using the CAS.upload() method.
It is possible to convert all of the data from a CAS table into a DataFrame by using the CASTable.to_frame() method, however, you do need to be careful. It will attempt to pull all of the data down regardless of size.
CASColumn objects work much in the same was as CASTable objects except that they operate on a single column of data like a pandas.Series.
|c[c.col > 1]||CASColumn||Series|
Pandas DataFrame vs. SASDataFrame¶
SASDataFrame is a subclass of pandas.DataFrame. Therefore, anything you can do with a pandas.DataFrame will also work with SASDataFrame. The only difference is that SASDataFrame objects contain extra metadata familiar to SAS users. This includes a title, label, name, a dictionary of extended attributes for information such as By groups and system titles (attrs), and a dictionary of column metadata (colinfo).
Also, since SAS has both formatted and raw values for By groups, SASDataFrame objects also have a method called SASDataFrame.reshape_bygroups() to change the way that By group information in represented in the DataFrame. See the By group documentation for more information.