Skip to content

Latest commit

 

History

History
277 lines (221 loc) · 12.9 KB

README.md

File metadata and controls

277 lines (221 loc) · 12.9 KB
Raw

Overview

This is a Python application that allows you to create/maintain/manage study configurations away from your implementations. experiment-server has several different interfaces (see below) to allow using it in a range of different scenarios. I've used it with Python, js and Unity projects. See the wiki for examples.

Documentation is available at https://shariff-faleel.com/experiment_server/

Content

Installation

Install it directly into an activated virtual environment:

$ pip install experiment-server

or add it to your Poetry project:

$ poetry add experiment-server

Usage

Configuration of an experiment

The configuration is defined in a toml file.

A config file can be generated as follows

$ experiment-server new-config-file new_config.toml

See example .toml below for how the configuration can be defined.

# The `configuration` table contains the settings of the study/experiment itself
[configuration]
# The `order` is an array of block names or an array of array of block names.
order = [["conditionA", "conditionB", "conditionA", "conditionB"]]
# The `groups` and `within_groups` are optional keys that allows you to define how the
# conditions specified in `order` will be managed. `groups` would dictate how the top 
# level array of `order` will be handled. `within_groups` would dictate how the conditions
# in the nested arrays (if specified) would be managed. These keys can have one 
# of the following values.
# - "latin_square": Apply latin square to balance the values.
# - "randomize": For each participant randomize the order of the values in the array.
# - "as_is": Use the order of the values as specified.
# When not specified, the default value is "as_is" for both keys.
groups = "latin_square"
within_groups= "randomize"
# The random seed to use for any randomization. Default seed is 0. The seed will be
# the value of random_seed + participant_index
random_seed = 0

# The subtable `variabels` are values that can be used anywhere when defining the blocks.
# Any variable can be used by appending "$" before the variable name in the blocks. See 
# below for an exmaple of how variables can be used
[configuration.variables]
TRIALS_PER_ITEM = 3

# Blocks are defined as an array of tables. Each block must contain `name` and the 
# subtable `config`. Optionally, a block can also specify `extends`, whish is a `name` of
# another block. See below for more explanation on how `extends` works

# Block: Condition A
[[blocks]]
name = "conditionA"

# The `config` subtable can have any key-values. Note that `name` and `participant_index`
# will be added to the `config` when this file is being processed. Hence, those keys 
# will be overwritten if used in this subtable.
[blocks.config]
trialsPerItem = "$TRIALS_PER_ITEM"
param1 = 1
# The value can also be a function call. A function call is represented as a table
# The following function call will be replaced with a call to 
# [random.choices](https://docs.python.org/3/library/random.html#random.choices)
# See `# Function calls` in README for more information.
param2 = { function_name = "choices", args = { population = [1 , 2 , 3 ], k = 2}}
param3 = { function_name = "choices", args = [[1 , 2 , 3 ]], params = { unique = true } }

# Block: Condition B
[[blocks]]
name = "conditionB"
extends = "conditionA"

# Since "conditionB" is extending "conditionA", the keys in the `config` subtable of 
# the block "conditionA" not defined in the `config` subtable of "conditionB" will be copied
# to the `config` subtable of "conditionB". In this example, `param1`, `param2` and 
# `trialsPerItem` will be copied over here.
[blocks.config]
param3 = [2]

See toml spec for more information on the format of a toml file.

The above config file, after being processed, would result in the following list of blocks for participant number 1:

[
  {
    "name": "conditionB",
    "extends": "conditionA",
    "config": {
      "param3": [
        2
      ],
      "trialsPerItem": 3,
      "param1": 1,
      "param2": [
        1,
        2
      ],
      "participant_index": 1,
      "name": "conditionB",
      "block_id": 0
    }
  },
  {
    "name": "conditionA",
    "config": {
      "trialsPerItem": 3,
      "param1": 1,
      "param2": [
        2,
        2
      ],
      "param3": [
        3
      ],
      "participant_index": 1,
      "name": "conditionA",
      "block_id": 1
    }
  },
  {
    "name": "conditionA",
    "config": {
      "trialsPerItem": 3,
      "param1": 1,
      "param2": [
        1,
        1
      ],
      "param3": [
        2
      ],
      "participant_index": 1,
      "name": "conditionA",
      "block_id": 2
    }
  },
  {
    "name": "conditionB",
    "extends": "conditionA",
    "config": {
      "param3": [
        2
      ],
      "trialsPerItem": 3,
      "param1": 1,
      "param2": [
        3,
        1
      ],
      "participant_index": 1,
      "name": "conditionB",
      "block_id": 3
    }
  }
]

Verify config

A config file can be validated by running:

$ experiment-server verify-config-file sample_config.toml

This will show how the expanded config looks like for the first 5 participants.

Loading experiment through server

After installation, the server can used as:

$ experiment-server run sample_config.toml

See more options with --help

The server exposes the following REST API:

  • [GET] /api/blocks-count / /api/blocks-count/:participant-id - Return the number of blocks in the configuration loaded. For a given config, the blocks-count will be the same for all participants.

  • [GET] /api/block-id / /api/block-id/:participant-id - Returns the current block-id. If participant-id is provided, the blcok-id of the participant will be returned, if not the default participant's block-id will be returned. Note that the block-id is 0 indexed. i.e., the first block's block-id is 0.

  • [GET] /api/active / /api/active/:participant-id - Returns the status for participant-id, if participant-id is not provided, will return the status of the default participant. Will be false if the participant was just initialized or the participant has gone through all blocks. To initialize the participant's status (or move to a given block), use the move-to-next or move-to-block endpoints.

  • [GET] /api/config / api/config/:participant-id - Return the config for participant-id, if participant-id is not provided, will return the config for the default participant.

  • [GET] /api/summary-data / /api/summary-data/:participant-id - Returns the summary of the configs for participant-id, if participant-id is not provided, returns the summary of the configs for the default participant. Currently, the summary is a JSON with the following keys

    • "participant_index"

    • "config_length"

  • [GET] /api/all-configs / /api/all-configs/:participant-id - Returns all the configs as a list for the participant-id, if participant-id is not provided, returns the configs for the default participant.This is akin having all the results from calling the config endpoint for each block in one list.

  • [GET] /api/status-string / /api/status-string/:participant-id - Returns status string for participant-id, if participant-id is not provided, returns statu string the default participant.

  • [POST] /api/move-to-next / /api/move-to-next/:participant-id - Move participant-id to the next block, if participant-id is not provided, move the default participant to the next block. If the participant was not initialized (active is false), will make be marked as active (active will be set to true). If the block the participant was in was the last block, they will be marked as not active (active will be set to false).

  • [POST] /api/move-to-block/:block-id / /api/move-to-block/:participant-id/:block-id - Move participant-id to the block number indicated by block-id, if participant-id is not provided, move the default participant to the block number indicated by block-id. If the participant was not initialized (active is false), will make be marked as active (active will be set to true). Will fail if the block-id is below 0 or above the length of the config.

  • [POST] /api/move-all-to-block/:block-id - Move all active participants (active returns true) to the block number indicated by block-id.

  • [POST] /api/shutdown - Shuts-down the server.

  • [PUT] /api/new-participant - Adds a new participant and returns the new participant-id. The new participant-id will be the largest current participant-id +1.

  • [PUT] /api/add-participant/:participant-id - Add a new participant with participant-id. If there is already a participant with the participant-id, this will fail.

For a Python application, experiment_server.Client can be used to access configs from the server. Also, the server can be launched programmatically using experiment_server.server_process which returns a Process object.

NOTE: If the config file served is changed, the new config will be loaded, but the state of the participants will be maintained. i.e., the added participants and the block id they are at will not change. To move the block ids for all active participants, you would have to call the move-all-to-block endpoint.

The server also provides a simple web interface, which can be accessed at / or /index. This interface allows to manage and monitor the flow of the experiment:

web UI screenshot

Loading experiment through API

A configuration can be loaded and managed by importing experiment_server.Experiment.

Generate expanded configs

A config file (i.e. .toml file), can be expanded to JSON with the following command

$ experiment-server generate-config-json sample_config.toml --participant-range 5

The above will generate the expanded configs for participant indices 1 to 5 as JSON output on stdout. This result can be written out to individual JSON files by setting the --out-dir/-d to a directory. See more options with --help

Function calls in config

A function call in the config is represented by a table, with the following keys

  • function_name: This should be one of the names in the supported functions list below.
  • args: The arguments to be passed to the function represented by function_name. This can be a list or a table/dict. They should unpack with * or ** respectively when called with the corresponding function.
  • (optional) params: function-specific configurations to apply with the function calls.
  • (optional) id: A unique identifier to group function calls.

A table that has keys other than the above keys would not be treated as a function call. Any function calls in different places of the config with the same id would be treated as a single group. Tables without an id are grouped based on their key-value pairs. Groups are used to identify how some parameters affect the results (e.g., unique for choices). Function calls can also be in configurations.variabels. Note that all function calls are made after the extends are resolved and variables from configurations.variabels are replaced.

Supported functions

  • choices: Calls random.choices. params can be a table/dictionary which can have the key unique. The value of unique must be true or false. By default unique is false. If it's true, within a group of function calls, no value from the population passed to random.choices is repeated for a given participant.

Example function calls

param = { function_name = "choices", args = [[1 , 2 , 3 , 4]], params = { unique = true } }
param = { foo = "test", bar = { function_name = "choices", args = { population = ["w", "x", "y", "z"], k = 1 } } }

For more on the experiemnt-server and how it can be used see the wiki

Wishlist (todo list?)

  • Improved docs
    • Add the option of using dict values in order