Skip to content

Latest commit

 

History

History
224 lines (184 loc) · 9.42 KB

README.md

File metadata and controls

224 lines (184 loc) · 9.42 KB
Raw

This is the code for an API which, given a to-do list, incentivizes tasks and schedules them. It will be used in future CompliceX experiments with Workflowy to-do lists.

Local usage

In order to be able to run the code, you need to have a mongo database.

After installing MongoDB, you should do the following steps:

  • Open a terminal window and start a connection with the database with the command mongod --dbpath <pathToDatabase> --port 27017
  • Open another terminal window and start the DB editor with the command mongo.
    • In order to create a user, initialize a collection etc., please refer to MongoDB's official websie.
    • Please initialize the following collections: log_dict, pr_transform, trees.
  • In the if __name__ == '__main__' part of the app.py file, the following info should be provided. Here is an example: 
    uri = "mongodb://ai4productivity:[email protected]/ai4productivity"
client = MongoClient(uri)
db = client["ai4productivity"]
collection = db["ai4productivity"]
  • Start the application by typing python app.py in the root folder of the API.
  • Send requests to the API. For this purpose, we recommend using Postman.

Sending requests

For successful communication with the API, a POST request should be sent via the following URLs with a pre-defined body structure (described in detail in the report).

The general URL for local testing looks like this: http://127.0.0.1:6789/api/<compulsoryParameters>/<additionalParameters>/tree/<userID>/<functionName>

The general URL for using the server online (on Heroku) looks like this: https://<HerokuAppCode>.herokuapp.com/api/<compulsoryParameters>/<additionalParameters>/tree/<userID>/<functionName>

Description of the URL parameters:

  • <method>: Method by which points are assigned
    • constant: Constant point assignment
    • length: Length heuristics
    • random: Random point assignment from a Normal distribution.
    • smdp: Calculates optimal points by using semi-Markov decision processes.
  • <scheduler>: Procedure by which tasks are scheduled
    • Schedulers for constant, length and random point-assignment methods:
      • basic: Basic scheduler
      • deadline: Deadline scheduler
    • Schedulers for smdp point-assignment method:
      • mdp: Method used by the SMDP incentivizing method.
  • <compulsoryParameters>
    • default_time_est: Default task time estimate (in minutes) to fill in if it is not provided by the user.
    • default_deadline: Default deadline (number of days, starting from today) to fill in if it is not provided by the user.
    • allowed_task_time: Time-estimation restriction for tasks, so that users do not enter long time estimations. If no restriction to impose is necessary, then the input should be inf.
    • min_sum_of_goal_values: Lower interval bound on the sum of goal values.
    • max_sum_of_goal_values: Upper interval bound on the sum of goal values.
    • min_goal_value_per_goal_duration: Lower interval bound on the ratio between a goal value and its duration (in minutes).
    • max_goal_value_per_goal_duration: Upper interval bound on the ratio between a goal value and its duration (in minutes).
    • points_per_hour: if 'true'-valued (true, t, 1), we assign points per hour. otherwise, we assign points for task completion.
    • rounding: The number of decimals to round to. For input of 0, the points will be rounded to the closest integer.
  • <additionalParameters>: (Differ for each method. Described in their own section.)
  • <userID>: Unique user identification code.
  • <functionName>: Type of request.
    • getTasksForToday: Outputs list of task for today.
    • updateTransform: Updates bias and scaling parameters for SMDP method.
    • updateTree: Updates the stored tree.

The additional parameters (<additionalParameters>) are dependent on the method that has been used (described below). Important: The order of all URL parameters is fixed!

You can use our URL generator to get the general URL to post to, before the last three parameters (userID, tree, and functionName).

Constant point-assignment point-assignment method (const)

  • Additional parameters
    • default_task_value: Constant value of points to be assigned to each task.
URL example: http://127.0.0.1:6789/api/constant/basic/30/14/inf/0/3000/0/60/t/2/10/tree/user123/getTasksForToday

<method>: constant
<scheduler>: basic
default_time_est: 30
default_deadline: 14
allowed_task_time: inf
min_sum_of_goal_values: 0
max_sum_of_goal_values: 3000
min_goal_value_per_goal_duration: 0
max_goal_value_per_goal_duration: 60
points_per_hour: t
rounding: 2
default_task_value: 10
<userID>: user123
<functionName>: getTasksForToday

Length heuristics point-assignment method (length)

  • There are no additional parameters for this method.
URL example: http://127.0.0.1:6789/api/length/deadline/30/14/inf/0/inf/0/inf/true/0/tree/user123/getTasksForToday

<method>: length
<scheduler>: deadline
default_time_est: 30
default_deadline: 14
allowed_task_time: inf
min_sum_of_goal_values: 0
max_sum_of_goal_values: inf
min_goal_value_per_goal_duration: 0
max_goal_value_per_goal_duration: inf
points_per_hour: true
rounding: 0
<userID>: user123
<functionName>: getTasksForToday

Random point-assignment method (random)

  • Additional parameters
    • distribution: The name of the probability distribution (according to NumPy) and their own parameters. So far, these distributions have been implemented:
      • uniform: Uniform distribution with parameters low (lower interval bound) and high (higher interval bound)
      • normal: Normal (Gaussian) distribution with parameters loc (mean value) and scale (standard deviation)
URL Example: http://127.0.0.1:6789/api/random/deadline/30/14/inf/0/10000/0/10/false/2/normal/1/100/tree/user123/getTasksForToday

<method>: random
<scheduler>: deadline
default_time_est: 30
default_deadline: 14
allowed_task_time: inf
min_sum_of_goal_values: 0
max_sum_of_goal_values: 10000
min_goal_value_per_goal_duration: 0
max_goal_value_per_goal_duration: 10
points_per_hour: false
rounding: 2
distribution: normal
loc: 1
scale: 100
<userID>: user123
<functionName>: getTasksForToday

SMDP point-assignment method (smdp)

  • Additional parameters
    • choice_mode: Mode of making time transitions while executing optimal policy
      • max: Choose the path that is most-likely to happen.
      • random: Make a random choice w.r.t. probabilities assigned to time transitions.
    • gamma: Discount factor float(0, 1)
    • loss_rate: Unit-time value that models cognitive effort float[0, inf)
    • num_bins: Number of time transitions int[1, inf)
    • planning_fallacy_const: Value that scales time estimates float(0, inf)
    • slack_reward: Reward associated with slack-off actions float[-inf, inf)
    • unit_penalty: Unit-time value that penalizes float[0, inf]
    • scale_type (optional): It represents the method by which points are scaled. If no scaling to be used, the inputting this parameter (and the scale_min and scale_max parameters) should be omitted.
      • no_scaling: Points are assigned according their pseudo-rewards (no change).
      • min_max: Points are assigned according to this formula task_reward = (task_reward - min_value) / (max_value - min_value) * (scale_max - scale_min) + scale_min.
      • mean_value: Points are assigned according to this formula task_reward = (task_reward - mean_reward) / (max_value - min_value) * (scale_max - scale_min) / 2 + ((scale_max + scale_min) / 2)
    • scale_min (optional): Lower interval bound. If inf, then the lower interval bound is not set.
    • scale_max (optional): Upper interval bound. If inf, then the upper interval bound is not set.

Notation:

  • [lower_bound, upper_bound]: closed interval
  • (lower_bound, upper_bound]: half-open interval
  • [lower_bound, upper_bound): half-open interval
  • (lower_bound, upper_bound): open interval
URL example: http://127.0.0.1:6789/api/smdp/mdp/30/14/inf/0/inf/0/inf/false/2/max/0.9999/1/2/1.39/0/0/min_max/1/2/tree/u123/getTasksForToday

<method>: random
<scheduler>: deadline
default_time_est: 30
default_deadline: 14
allowed_task_time: inf
min_sum_of_goal_values: 0
max_sum_of_goal_values: inf
min_goal_value_per_goal_duration: 0
max_goal_value_per_goal_duration: inf
points_per_hour: false
rounding: 2
choice_mode: max
gamma: 0.9999
loss_rate: 1
num_bins: 2
planning_fallacy_const: 1.39
slack_reward: 0
unit_penalty: 0
scale_type: min_max
scale_min: 1
scale_max: 2
<userID>: user123
<functionName>: getTasksForToday

Potential issues

If you encounter any problem related to the API, please submit a GitHub issue.

Required Python Packages

All required Python packages are listed in the requirements.txt file.

Citation

If you use this code in academic work, please cite the report:

@article{consul2021optimal,
  title={Optimal To-Do List Gamification for Long Term Planning},
  author={Consul, Saksham and Stojcheski, Jugoslav and Felso, Valkyrie and Lieder, Falk},
  journal={arXiv preprint arXiv:2109.06505},
  year={2021}
}
}

Acknowledgements

This project uses code from: