All notable changes to this project will be documented in this file. This project adheres to Semantic Versioning starting with version 0.0.1a1
- fixed a bug that prevented toggling the
secure_url
option via.env
- added the support for toggle between
HTTP
andHTTPS
for securingbot URL
when explaining models viaREST API
- redefined file io constants for source code clarity
- fixed a bug that caused errors in retrieving models due to an unexpected
reverse
argument
- enabled
data_source_path
argument passing inprocess queue
- explanations UI improvements
- refined constant definitions
- fixed version inconsistency in the UI by fetching the version directly from the backend configurations
- fixed frontend doc url inconsistencies by defining them in the backend
- fixed api url inconsistencies by defining them in the backend
- removed custom dir and moved dime_diet_classifier to a dir called 'dime_components' in init dir list
- improved favicon support and the manifest
- UI Improvements
- removed redundant .env file loading
- turned off version check for config file
- added Sinhala-English code-switching typing support
- fixed process queue initialization exception on linux-based devices
- optimized react js codebase for
DIME Server
- renamed
process
toprocess_queue
and added support to specify a specificdata_source_path
to preserve the process queue - moved default in memory process queue to
dime_cache
- fixed model count error after changing the active model path in
DIME Server
- fixed
dime init
bug where.dime_cache
was not properly copied to the project directory
- updated package dependencies to avoid conflicts with rasa dependencies
- dependency updates for
pandas
andnumpy
to resolve conflicts on Google Co-Lab - extended DIME support for
Python 3.7
to enable DIME on Co-Lab notebooks - updated
dime init
dir structure to include DIME-compatiblecustom DIET classifier
component - updated init dir instructions and readmes
- added an updated initial rasa model taken from rasa init trained using custom DIET classifier
- moved
.env
files from server root to project root - added a dummy env file to the init project root
DIME CLI
supportsquiet
mode which suppresses all logs and only outputs stdout and stderr for final outputs for the explanation generation process- added full React support
- added
Download-able
DIME Explanations - added
Upload
feature to theDIME Server
which allows to upload previously generated dDIME explanation JSON files to the server - added
Peak
feature to theDIME Server
which allows to quickly visualize dime explanation files without having to upload it to the server - DIME Server allows to manage both
models
andexplanations
available on the server, locally - added a
In-Memory Process Queue
to allow users toAbort
explanation generation requests made from the server - added support for
Rasa 2.8.8
models - added the ability to load
custom rasa pipeline components
while loading a model for explanation generation. (only available on local models) - added
--quiet
flag todime init
which creates a new DIME project from scratch without prompting for user inputs for project location verification
- CLI-related docstrings were added for clarity based on the Google Python Styleguide
- added the ability to track explanation generation requests sent from DIME server using a unique request ID attached to the process IDs. Aborting requests was not supported in previous versions
- server cache was disabled and the behavior was assigned to process queue component of the DIME server
- removed React browser router and utilized hash router instead to avoid root route conflicts and routing loops
softmax
function andexp_norm_softmax
functions are deprecated and will be removed in Rasa 2.0.0 onwards.to_probability_series
can be utilized to convert feature importance score series to a probability seriesAccuracy
andF1-Score
are deprecated as metrics in DIME core.Model Confidence
should be used as the default metric for all feature importance calculations
- fixed a JSON serialization error occurred while trying to parse numpy values given by the normalization, probability series calculation and feature selection methods
DIME sever
is now able to serve react frontends and static files, and able to redirect to the correct hash route when non hash routes are being calledDIME server
is able to correctly process subprocess stdout and stderr byes and figure out DIME exceptions while generating explanations- DIME config validations were refined due to not being able to detect empty data instances specified in the DIME configuration file
- React logs were suppressed in production environments
- server-sent configuration validations were properly implemented