Examples
Several examples are included to help you become familiar with Pyke. These are all in an examples directory:
$ cd examples/towers_of_hanoi $ python >>> import driver >>> driver.test(2) got 1: ((0, 1), (0, 2), (1, 2)) got 2: ((0, 2), (0, 1), (2, 0), (1, 2), (0, 2))
Each example is in its own sub-directory and has a README.txt file to get you started. They all have .krb files and a Python module to run the example that also demonstrates how to call Pyke from your Python program.
Family_relations
This is a very good basic example to start with.
The family_relations example takes an initial set of facts about people (stated in a .kfb file):
son_of(david_r2, david_r, sarah_r) daughter_of(shirley, david_r, sarah_r)
And figures out how any two people are related:
david_r2, shirley are ('brother', 'sister')
This same problem is solved in four different ways so that you can compare them:
- Forward-chaining only
- Backward-chaining only
- Backward-chaining only with a few rule optimizations that make the rules run 100 times faster!
- A mix of forward-chaining and backward-chaining with some use of plans added too.
The driver.py program also demonstrates how to use krb_traceback and the print_stats function.
Knapsack
At the PyCon 2008 conference, somebody asked about the knapsack problem. We found a solution in Prolog here (starting on page 19), and rewrote it in Pyke. This is a quick simple little example.
Sqlgen
Pyke was originally developed as the control component for a web framework. This example shows how Pyke can automatically generate SQL SELECT statements, given a set of tables that the calling program has keys to and a tuple of the desired column names. Column names specified at the top-level in this tuple are expected to have a single value each. Nested tuples are used when multiple rows are expected. The column names in nested tuples make up the columns in the result rows.
The top-level goal returns a plan that takes the key values for the initial set of tables given to the goal and returns an immutable dictionary mapping the column names to the values retrieved from the database. The plan may be used repeatedly without re-running the rules each time to figure out the SELECT statements. Thus, this acts like a SELECT statement compiler resulting in queries with virtually no extra overhead. It is not, however, an Object Relational Mapper (ORM).
The data model used for the requested columns is that tables inherit the columns from tables they link to. So if there is a 1-many relationship between tables A and B (1 A row has many B rows), the B table inherits the columns from the A table through it's link to table A. The Pyke rules will automatically figure out the table joins for this.
The program automatically introspects the schema information. For this example, it assumes that id is the primary key for each table, and that when one table links to another, it uses the target table name suffixed with _id as the column name.
This example was originally done using MySQL and includes the .sql files to create the database, tables, and example data. The example has since been converted to use the Sqlite3 database to make it easier to run, as Sqlite3 does not require any setup (the Sqlite3 database file is included in the example).
Sqlgen lacks more general capabilities that would be required for real use, but may serve as a starting point for another project that's more complete.
This example also has much more elaborate rules than the prior two examples and is a very real example of generating plans.
Web_framework
This example completes the Python web framework demo by adding rules to automatically generate code to render HTML templates from the HTMLTemplate package (you can run pip install HTMLTemplate or easy_install HTMLTemplate to install the HTMLTemplate package). This example uses the sqlgen example, above, to generate the SQL statements.
An HTMLTemplate does not include anything resembling program code in it, so that your graphics designers can completely own the html files without the developers having to modify them in any way.
Note that the code generated here is fully cooked code, custom built for that specific schema and HTML template. This runs extremely fast because there is nothing left at run-time concerning parsing and figuring out the HTML template, or constructing the SQL statements.
A test was done comparing this web framework example to the same example done in TurboGears 2 running against the same MySQL database. The results of the siege benchmark tests show that Pyke is just over 10 times faster than TurboGears 2:
- Pyke: 791 trans/sec - TurboGears 2: 76 trans/sec
The demo is packaged as a WSGI application. It also demonstrates the use of multiple rule bases by using the sqlgen example above, as well as the caching and reuse of plans to achieve the order of magnitude improvement in performance over current practice.