|
|
|
Here scheme of the project structure |
|
|
\ No newline at end of file |
|
|
|
# GLOBALS
|
|
|
|
DATABASE: SQLite
|
|
|
|
|
|
|
|
SAMPLE_SIZE=442 (max = 442 due to size of train dataset);<br>
|
|
|
|
CASE_ID=0 stands for geological dataset, CASE_ID=1 - for social dataset.<br>
|
|
|
|
|
|
|
|
PREPROCESSING (in this version it remains constant and not available for user to tune)<br>
|
|
|
|
First, we encode discrete data, then discretize continuous data. <br>
|
|
|
|
_The purpose of this action and more details can be found in section "About BAMT algorithms"._<br>
|
|
|
|
encoder = pp.LabelEncoder()<br>
|
|
|
|
discretizer = pp.KBinsDiscretizer(n_bins=5, encode='ordinal', strategy='quantile')<br>
|
|
|
|
|
|
|
|
EXAMPLE PARAMS <br>
|
|
|
|
Geological dataset:<br> has_logit=0,<br> use_mixture=0,<br> score_function="K2"<br><br>
|
|
|
|
Social dataset:<br> has_logit=1,<br> use_mixture=1,<br> score_function="MI" (different params to avoid isolated vertices)
|
|
|
|
|
|
|
|
# API
|
|
|
|
There are 4 main and 1 additional modules implemeneted. Additional modules are tests, we use them only in dev stage.<br>
|
|
|
|
Each module follows this pattern:<br>
|
|
|
|
1. Controller. File with query
|
|
|
|
2. Service. File with core functions
|
|
|
|
3. Models. File with declarations of tables in database
|
|
|
|
4. Schema. File with docs.<br><br>
|
|
|
|
We have ordered the modules accoring their usage in main scenario.
|
|
|
|
|
|
|
|
## AuthMod
|
|
|
|
This module provides a communication between user and auth system.
|
|
|
|
### Controller
|
|
|
|
get_token(email, password): get token by email and password;<br>
|
|
|
|
signup(email, password): registrate user in database.<br>
|
|
|
|
### Models
|
|
|
|
Declare the tables related to auth system.
|
|
|
|
### Service
|
|
|
|
Here we defined functions to work with auth system.
|
|
|
|
## Example
|
|
|
|
Demonstarte an instanse of pretrained network.
|
|
|
|
### Controller
|
|
|
|
get_example(case_id): return a network pretrained with
|
|
|
|
|
|
|
|
## Experiment
|
|
|
|
One of the most important module in applictaion. It is responsible for training bayssian network, sample from it.
|
|
|
|
### Controller
|
|
|
|
get(owner, name, case_id, bn_params): validate input, train network, sample from it and save data.<br>
|
|
|
|
IMPORTANT NOTE: in this version we have to unify format of sample (_actually, there are 2 types: discrete and continuous_).
|
|
|
|
We descritize continouos data, bins equal to output of `np.histogram_bin_edges()` on train dataset for better comparison.<br>
|
|
|
|
get_root_nodes(case_id): return initial nodes from dataset.<br>
|
|
|
|
### Models
|
|
|
|
Declare tables with networks and samples.
|
|
|
|
### Service
|
|
|
|
Core functions to fit bayessian network and save them.
|
|
|
|
## bn_manager
|
|
|
|
Module provides operations with bayessian networks in database such as: find BN(-s) if exists, delete, put and train.
|
|
|
|
### Controller
|
|
|
|
get_BN(owner): return a list of bn(-s) (**with data about them**);<br>
|
|
|
|
get_BN_names(owner): return a list of bn(-s) names user owns;<br>
|
|
|
|
get_sample(owner, name, node): return an sample data array with size=SAMPLE_SIZE;<br>
|
|
|
|
remove(owner, name): remove bn (and its sample) from database.<br>
|
|
|
|
|
|
|
|
|
|
|
|
Scheme of the project structure
|
|
|
|
 |
