decision-making

When you need to compare and decide which apartment to buy, which guitar suits you best, which candidates you should consider for the job, what is the perfect job for you, etc.

A software framework to enable human decision making

• Based on socratic model
• Simple user interface that guides the user by asking questions
• Parts: model building, data building, data analysis

Workflow:

0. Workspace preparation, for data storage
   1. Create folder decision-making/data
   2. Create folder decision-making/data/dataset
   3. Create folder decision-making/data/model
1. Model building:
   1. Name the model
   2. Add parameters: select type, name, additional arguments (such as value range), normalizer
   3. Optionally, add weights to the parameters
   4. Optionally, create or link existing datasets to the model
   5. Store the model
2. Data building:
   1. Gather data
   2. Add data points to datasets
   3. Evaluate (and validate) the model
3. Data analysis:
   1. Filter data
   2. Sort data
   3. Plot data
   4. Chart data
   5. Remove outliers

Glossary:

• Model: Defines the object of decision-making (apartment, interview candidate, etc)
• Parameter: Describes the object of decision-making (cost, color, year). Each parameter describes a value and computes a score (0-100). The score represents the parameter's normalized value. Note: the numerical values are stored in ParameterDatas which are stored in DataPoints. Parameters only description and computation.
• Normalizer: A function that transforms a parameter value into a score. Ex. an apartment buyer could consider only apartments built after the 1960s. The simplest normalizer would be a step, which scores all years before 1960 to 0, and all years after 1960 to 100. A more complex one could be a positive linear step, which maps, for instance, all years before 1960 to zero, all years from 1960 to 2000 to the interval 0-100, and all years after 2000 to a score of 100
• Score: A parameter's value, normalized in the interval of 0-100
• Dataset: A set of data points that belong to a model
• Data point: One ParameterData for each parameter of the model. The decision candidate. This represents, for example, one apartment, one interview candidate, etc.
• ParameterData The name, numerical value and score stored in the dataset, that correspond to a model parameter

TODO: rename Normalizer into Scoring / Scorer / ScoreFunction / ScoringFunction. It is a more intuitive metaphor

---

Model building

• The model describes the setup for the decision-making problem
• The model consists of several parameters that aid in decision making
• The model is created by the user
• The model is stored in a binary file
• The datasets of the model are stored in json files

TODO

• Model can be based on SWOT
• All parameters can be distributed on a circle with radius of 100
• Parameters can be normalized automatically by relative comparison. Or absolute comparison can be made by user
• ADD TESTS (unit, integration)

Data building

• For a chosen (created) model, data is entered

Data analysis

TODO

• Radar chart is created
• Filtering of data points based on parameter (ascending, descending, outlier)

---

User interface

CMD

Available commands

Create a model: ./cli.py model --create

Delete a model: ./cli.py model --name <name> --delete

Describe the model: ./cli.py model --name <name> --describe

Change the model name: ./cli.py model --name <name> --rename

Delete a model parameter: ./cli.py model --name <name> --delete-param <param_name>

Add a model parameter: ./cli.py model --name <name> --add-param

Modify parameter weight: ./cli.py model --name <name> --weight-param <param_name> <new_weight>

List (describe) all parameters: ./cli.py parameters --list

List (describe) all normalizers: ./cli.py normalizers --list

Add dataset to model ./cli.py model --name <name> --add-dataset

List datasets of model ./cli.py model --name <name> --list-datasets

Delete dataset from model ./cli.py model --name <name> --delete-dataset

Evaluate datasets of model ./cli.py model --name <name> --evaluate-dataset

Model building TODO

• Add tests
• Parameters should be storable just like datasets
• Draw normalizer example
• Fully automated add-param for scripting

Data building TODO

• Add datapoint to dataset ./cli.py dataset --name <name> --datapoint-add <name>
• Rename datapoint of dataset ./cli.py dataset --name <name> --datapoint-rename <name>

GUI

TODO

---

Use cases

A selection of use cases to demonstrate the desired capabilities of the final product:

• Apartment buyer
• Car buyer
• Recruiter
• SWOT analysis (manager)
• Scrum Product Owner (feature prioritization)

---

The business model

Python backend + (tbd) frontend -> web-based solution. Thus, the business model is software-as-a-service, combined with fee-for-service (in future). The user creates an online account, selecting one of the account types from below:

Account type	Models	Parameters	Datasets per model	Datapoints per dataset	Access to model library	Access to AI-based data analysis tools	Expires in
Free	1	5	1	10	❌	❌	30 days
Basic	5	10	5	20	❌	❌	1 year
Full	10	100	10	1000	✔️	✔️	1 year

---

Example

• Open-loop model: Buying an apartment
• Closed-loop model: deciding each day how many hours to watch tv, how many hours to exercise, etc

Buying an apartment

Model

Parameters

Parameters are normalized by relating each data entry. Ex after entering data for 5 apartments, min and max are converted from 0 to 100. All the others come in between.

• Price
• Area sqm
• Year of construction
• Vastike
• Floor
• Zone
• Distance to closest mall
• (Distance to school)
• (Distance to recreational areas)

---

Future stuff

Closed-loop model

• Optimizable via Kalman-filter approach

Open-loop model

• Can it be transformed into a closed-loop model ? That means, finding data about similar decisions taken by others in the past
• Can generate a poll if convertible to closed-loop model

• The model can be open-loop (once a decision is made, it cannot be repeated, eg. buying an apartment)
• The model can be closed-loop (decisions can be repeated)