Example of a D-efficient RUM design with conditions in ChoiceDesign

This notebook illustrates how to use ChoiceDesign to generate a D-efficient experimental design with conditions for a Random Utility Maximisation (RUM) model. Given a set of attributes and prior parameters, ChoiceDesign uses a variation of the random swapping algorithm [1] to minimise the D-error of the information matrix of a Multinomial Logit (MNL) model.

Step 1: Load modules, define design parameters and set attributes

The following lines load:

  • EffDesign: the class of efficient designs,

  • Attribute and Parameter: the classes of attributes and parameters, respectively.

[1]:
from choicedesign.design import EffDesign
from choicedesign.expressions import Attribute, Parameter

Each attribute is defined by the Attribute class. The arguments of this class are:

  • name: a string with the attribute name,

  • levels: a list of levels of the attribute,

Each attribute is alternative-specific. Hence, attributes must be defined for each alternative that contains them.

The following lines define 2 alternatives, named alt1 and alt2, and 4 attributes named from \(A\) to \(D\):

[2]:
alt1_A = Attribute('alt1_A',[1,2,3])
alt1_B = Attribute('alt1_B',[10,15,15.5])
alt1_C = Attribute('alt1_C',[0,3,5])
alt1_D = Attribute('alt1_D',[0,1,2])

alt2_A = Attribute('alt2_A',[1,2,3])
alt2_B = Attribute('alt2_B',[10,15,15.5])
alt2_C = Attribute('alt2_C',[0,3,5])
alt2_D = Attribute('alt2_D',[0,1,2])

Step 2: Construct efficient design object and generate initial design matrix

The second step consists of constructing the experimental design object, which requires the following parameters:

  • X: A list of Attribute class elements,

  • ncs: The number of choice situations.

The following lines define a object named design using EffDesign of 16 choice situations:

[3]:
design = EffDesign(
    X = [alt1_A,alt1_B,alt1_C,alt1_D,
         alt2_A,alt2_B,alt2_C,alt2_D],
    ncs=18)

After the design object is defined, the method gen_initdesign() generates the initial design matrix. This method accepts the following optional parameters:

  • cond: List of conditions that the final design must hold. Each element is a string that contains a single condition. Conditions can be of the form of binary relations (e.g., X > Y where X and Y are attributes of a specific alternative) or conditional relations (e.g., if X > a then Y < b where a and b are values). Users can specify multiple conditions when the operator if is defined, separated by the operator &.

  • seed: Random seed

We will define three conditions for this design. The following lines define the conditions and integrate them into the initial design matrix:

[4]:
cond = ['alt1_A > alt2_A',
        'if alt1_B > 10 then alt2_A < 3',
        'if alt1_A > 1 then alt2_A < 3'
        ]
[5]:
init_design = design.gen_initdesign(cond=cond)
init_design
[5]:
alt1_A alt1_B alt1_C alt1_D alt2_A alt2_B alt2_C alt2_D
0 2.0 10.0 0.0 2.0 1.0 10.0 5.0 0.0
1 3.0 15.0 3.0 1.0 1.0 15.0 5.0 0.0
2 2.0 10.0 5.0 2.0 1.0 15.5 0.0 0.0
3 2.0 10.0 3.0 1.0 1.0 10.0 5.0 1.0
4 3.0 15.5 3.0 2.0 1.0 15.0 3.0 0.0
5 2.0 10.0 0.0 0.0 1.0 15.0 3.0 1.0
6 2.0 15.5 5.0 2.0 1.0 10.0 3.0 1.0
7 2.0 10.0 5.0 0.0 1.0 15.0 0.0 2.0
8 3.0 10.0 0.0 2.0 2.0 10.0 5.0 2.0
9 2.0 10.0 5.0 2.0 1.0 10.0 3.0 0.0
10 2.0 10.0 0.0 1.0 1.0 15.0 5.0 1.0
11 3.0 15.5 3.0 0.0 2.0 15.5 0.0 1.0
12 3.0 10.0 5.0 1.0 1.0 10.0 5.0 0.0
13 3.0 15.5 5.0 1.0 1.0 15.5 0.0 2.0
14 2.0 15.5 5.0 2.0 1.0 10.0 5.0 2.0
15 3.0 15.5 5.0 2.0 1.0 10.0 3.0 2.0
16 3.0 10.0 5.0 2.0 1.0 15.0 0.0 1.0
17 2.0 15.0 5.0 0.0 1.0 15.5 0.0 0.0

Step 3: Set the utility functions

ChoiceDesign uses a native expression system to define utility functions. Parameters and attributes are combined using standard arithmetic operators. For this, we use the Parameter class, which requires the following arguments:

  • name: The parameter name

  • prior: The prior value

The following lines define four parameters:

[6]:
beta_A = Parameter('beta_A',-0.1)
beta_B = Parameter('beta_B',-0.02)
beta_C = Parameter('beta_C',0.1)
beta_D = Parameter('beta_D',0.15)

Then, the utility functions are defined using standard arithmetic operators. We will assume a linear utility function for each alternative.

[7]:
V1 = beta_A * alt1_A + beta_B * alt1_B + beta_C * alt1_C + beta_D * alt1_D
V2 = beta_A * alt2_A + beta_B * alt2_B + beta_C * alt2_C + beta_D * alt2_D

The utility functions must be stored in a dictionary object. In this dictionary, each key is a consecutive number from 1 to the number of alternatves. The values of each key are the corresponding utility functions:

[8]:
V = {1: V1, 2: V2}

Step 3: Optimise the initial design, given the utility functions and priors:

The method optimise() starts the D-error minimisation routine, given the initial design matrix and the utility functions. This method requires the following parameters:

  • init_design: The objective design matrix to optimise

  • V: The dictionary object with utility functions

  • model: The base model of the efficient design. By default is 'mnl' for a Multinomial Logit model.

In addition, optimise() admits the following optional parameters:

  • iter_lim: number of iterations before the algorithm stops.

  • noimprov_lim: Number of iterations without improvement before the algorithm stops,

  • time_lim: time (in minutes) before the algorithm stops,

  • seed: Random seed

  • verbose: Whether status messages and progress are shown.

The outputs of optimise are:

  • optimal_design: The optimised design matrix

  • init_perf: The initial D-Error

  • final_perf: The D-error of the last stored design

  • final_iter: The last iteration number

  • ubalance_ratio: The utility balance ratio. A 0% value indicates strict dominance of an alternative, whereas 100% indicates equal market shares.

The following line starts the optimisation routine during 1 minute:

[9]:
optimal_design, init_perf, final_perf, final_iter, ubalance_ratio = design.optimise(init_design=init_design,V=V,model='mnl',time_lim = 1, verbose = True)
Evaluating initial design
Optimization complete 0:00:59 / D-error: 0.034146
Elapsed time: 0:01:00
D-error of initial design:  0.052387
D-error of last stored design:  0.034146
Utility Balance ratio:  95.14 %
Algorithm iterations:  46180

Blocking the design

The optimal design can be blocked using the method gen_blocks(). This method randomly creates candidate blocks and keeps the one with the minimum correlation between the blocking column and all the attributes. The method allows for the following arguments:

  • optimal_design: the experimental design

  • n_blocks: number of blocks.

  • n_iter (optional): number of iterations of the blocking algorithm

The following line creates 4 blocks in the optimal design:

[10]:
optimal_design_blocked = design.gen_blocks(optimal_design,n_blocks=3)

Lastly, the optimal design can be printed:

[11]:
optimal_design
[11]:
CS alt1_A alt1_B alt1_C alt1_D alt2_A alt2_B alt2_C alt2_D Block
0 1.0 3.0 10.0 0.0 2.0 1.0 15.0 5.0 1.0 1
1 2.0 3.0 10.0 3.0 0.0 2.0 15.0 5.0 2.0 2
2 3.0 2.0 10.0 5.0 1.0 1.0 15.5 0.0 1.0 3
3 4.0 3.0 15.5 5.0 0.0 1.0 10.0 3.0 2.0 1
4 5.0 2.0 15.5 5.0 2.0 1.0 10.0 3.0 0.0 2
5 6.0 2.0 15.0 0.0 1.0 1.0 10.0 5.0 1.0 3
6 7.0 3.0 15.5 3.0 2.0 1.0 10.0 5.0 0.0 1
7 8.0 3.0 10.0 3.0 0.0 1.0 15.0 3.0 2.0 1
8 9.0 3.0 10.0 5.0 2.0 2.0 15.5 3.0 0.0 1
9 10.0 2.0 15.5 5.0 2.0 1.0 10.0 3.0 0.0 3
10 11.0 2.0 10.0 0.0 2.0 1.0 15.5 5.0 0.0 2
11 12.0 3.0 15.5 3.0 1.0 1.0 10.0 5.0 2.0 1
12 13.0 2.0 10.0 0.0 2.0 1.0 15.5 5.0 0.0 2
13 14.0 3.0 10.0 5.0 0.0 1.0 15.0 0.0 2.0 2
14 15.0 2.0 10.0 5.0 1.0 1.0 15.0 0.0 1.0 3
15 16.0 2.0 15.5 5.0 1.0 1.0 10.0 0.0 1.0 2
16 17.0 2.0 10.0 5.0 2.0 1.0 15.0 0.0 0.0 3
17 18.0 2.0 15.0 5.0 2.0 1.0 10.0 0.0 1.0 3

(optional) Evaluate the design

The method evaluate() allows to evaluate a design stored in a data frame, under the specification provided when EffDesign was initialised. evaluate() requires the following parameters:

  • optimal_design: The objective design matrix to evaluate

  • V: The dictionary object with utility functions

  • model: The base model of the efficient design. By default is mnl for a Multinomial Logit model.

[12]:
perf, ubalance = design.evaluate(optimal_design,V,model='mnl')

print(perf, ubalance)
0.034145726914017696 95.1394849419413

References

[1] Quan, W., Rose, J. M., Collins, A. T., & Bliemer, M. C. (2011). A comparison of algorithms for generating efficient choice experiments.