Skip to content

Latest commit

 

History

History
153 lines (108 loc) · 7.06 KB

README.md

File metadata and controls

153 lines (108 loc) · 7.06 KB
Raw

Orca_test

This is a library of assertions about the characteristics of tables, columns, and injectables that are registered in Orca.

The motivation is that UrbanSim model code expects particular tables and columns to be in place, and can fail unpredictably when data is not as expected (missing columns, NaNs, negative prices, log-of-zero). These failures are rare, but hard to debug, and can happen at any time because data is modified as models run.

Orca_test assertions can be included in model steps or used as part of the data preparation pipeline. The goal for this library is for it to be useful (1) as a model development aid, (2) for exception handling as simulations run, and (3) for documenting the data specs required by different UrbanSim templates.

Installation

Clone this repo and run python setup.py develop. Won't be of much use without Orca and some project that's using it for simulation orchestration.

Usage

You can either make assertions directly by calling individual orca_test functions, or assert a full set of characteristics at once. These characteristics are expressed as nested python classes (similar to sqlalchemy), and in the future will have an equivalent YAML syntax.

If an assertion passes, nothing happens. If it fails, an OrcaAssertionError is raised with a detailed message. Orca_test is written to be as computationally efficient as possible, and the main cost will be the generation of tables or columns that have not yet been cached.

Assertions are chained as necessary: for example, asserting a column's minimum value will automatically assert that it is numeric, that missing values are coded in a particular way (np.nan by default), that the column can be generated without errors, and that it is registered with orca.

Example

import orca_test as ot
from orca_test import OrcaSpec, TableSpec, ColumnSpec

# Define a specification
o_spec = OrcaSpec('my_spec',

	TableSpec('buildings', 
		ColumnSpec('building_id', primary_key=True),
		ColumnSpec('residential_price', min=0, missing=False)),

	TableSpec('households',
		ColumnSpec('building_id', foreign_key='buildings.building_id', missing_val_coding=-1)),
	
	TableSpec('residential_units', registered=False),
	
	InjectableSpec('rate', greater_than=0, less_than=1))

# Assert the specification
ot.assert_orca_spec(o_spec)

Working demos

  • development_tests.py in this repo
  • In the ual-development branch of UAL/bayarea_urbansim, the model steps include orca_test assertions to validate expected data characteristics (ual.py)

API Reference

There's fairly detailed documentation of individual functions in the source code.

Classes

  • OrcaSpec( spec_name, optional TableSpecs, optional InjectableSpecs )
  • TableSpec( table_name, optional characteristics, optional ColumnSpecs )
  • ColumnSpec( column_name, optional characteristics )
  • InjectableSpec( injectable_name, optional characteristics )
  • OrcaAssertionError

Asserting sets of characteristics

  • assert_orca_spec( OrcaSpec ) -- asserts the entire nested spec
  • assert_table_spec( TableSpec )
  • assert_column_spec( table_name, ColumnSpec )
  • assert_injectable_spec( InjectableSpec )

Table assertions

Argument in TableSpec() Equivalent low-level function
registered = True assert_table_is_registered( table_name )
registered = False assert_table_not_registered( table_name )
can_be_generated = True assert_table_can_be_generated( table_name )

Column assertions

Argument in ColumnSpec() Equivalent low-level function
registered = True assert_column_is_registered( table_name, column_name )
registered = False assert_column_not_registered( table_name, column_name )
can_be_generated = True assert_column_can_be_generated( table_name, column_name )
numeric = True assert_column_is_numeric( table_name, column_name )
missing_val_coding = np.nan, 0, -1 assert_column_missing_value_coding( table_name, column_name, missing_val_coding )
missing = False assert_column_no_missing_values( table_name, column_name, optional missing_val_coding )
max_portion_missing = portion assert_column_max_portion_missing( table_name, column_name, portion, optional missing_val_coding )
primary_key = True assert_column_is_primary_key( table_name, column_name )
foreign_key = 'parent_table_name.parent_column_name' assert_column_is_foreign_key( table_name, column_name, parent_table_name, parent_column_name, optional missing_val_coding )
max = value assert_column_max( table_name, column_name, max, optional missing_val_coding)
min = value assert_column_min( table_name, column_name, min, optional missing_val_coding )

Notes

Providing a missing_val_coding in a ColumnSpec() indicates that there should be no np.nan values in the column. Assertions involving a min, max, or max_portion_missing will take into account the missing_val_coding that's been provided.

For example, asserting that a column with values [2, 3, 3, -1] has min = 0 will fail, but asserting that it has
min = 0, missing_val_coding = -1 will pass.

Injectable assertions

Argument in InjectableSpec() Equivalent low-level function
registered = True assert_injectable_is_registered( injectable_name )
registered = False assert_injectable_not_registered( injectable_name )
can_be_generated = True assert_injectable_can_be_generated( injectable_name )
numeric = True assert_injectable_is_numeric( injectable_name )
greater_than = value assert_injectable_greater_than( injectable_name, value )
less_than = value assert_injectable_less_than( injectable_name, value )
has_key = str assert_injectable_has_key( injectable_name, str )

Development wish list

  • Add support for specs expressed in YAML
  • Write unit tests and set up in Travis
  • Make compatible with python 3

Sample YAML syntax (not yet implemented)

- orca_spec:
  - name: my_spec
  
  - table_spec:
    - name: buildings
    - column_spec:
      - name: building_id
  	  - primary_key: True
    - column_spec:
  	  - name: residential_price
  	  - min: 0
  	  - missing: False
  
  - table_spec:
    - name: households
    - column_spec:
  	  - name: building_id
  	  - foreign_key: buildings.building_id
  	  - missing_val_coding: -1
  
  - table_spec:
    - name: residential_units
    - registered: False
    
  - injectable_spec:
    - name: rate
    - greater_than: 0
    - less_than: 1