richard-churchman
5f6bbe45f7
Major bug fixes relating to the Redis implementation in Reprocessing and Search Key Cache were identified and fixed as part of the rekeying development. Included functionality to stop the Exhaustive training. Fixed the Unsupervised learning functionality in Exhaustive Training. Embedded the Mock Data CSV data into a class to make it compatible with Docker without needing to modify the Dockerfile (i.e, build). Included a new table that stores training data to enhance support capabilities for Exhaustive. Reintroduced and validated the Multitenancy functionality. Postgres is now fully deprecated in place of Redis for cache, and all implementations of the Interfaces for Postgres have been removed. No migrations exist to support this deprecation, and Postgres as a cache is now wholly unsupported. Removed the Indexing background thread, as Redis does not require it, given that it is a key-value pair database. Created a RedisCommandFlag Environment Variable to support Fire and Forget write to Redis in Invocation. Implemented functionality to Export, Peek and Download configuration environments using a secure, encrypted scheme. Created a permission to access the functionality via the user interface. Extensive database-based logging is included to provide for a comprehensive audit. Created functionality to Import the encrypted Export file inside a database transaction to ensure consistency. As Docker is now the primary deployment model, log4net instantiation has been moved to an approach that primarily uses programmatic instantiation of the ConsoleLogger and RollingFileAppender given Environment Variables. Implemented functionality to support remote \ mounted Log4Net XML Serialisation files for more advanced log4net configurations. Reintroduced access to counters via the user interface and created a permission. Took the opportunity to implement ReSharper linter suggestions. Disabled the Lock button across the whole application, with that functionality being reserved for future use. It is intended that locking happens programmatically on the basis of dependencies in future commits. Fixed several significant end-use bugs which became apparent during testing of the refactored code relating to case management and visualisations. Fixed certain other minor bugs that came up during testing. Brought documentation up to date for this commit. Brought the Docker Compose file up to date for this commit. Removed Redundant Environment Variables for this commit. Made the README file less verbose and unnecessary.
2.0 KiB
layout, title, nav_order, parent, grand_parent
| layout | title | nav_order | parent | grand_parent |
|---|---|---|---|---|
| default | Exhaustive Adaptation Training Variable Statistics | 3 | Exhaustive Adaptation | Configuration |
🚀Speed up implementation with hands-on, face-to-face training from the developer.
Exhaustive Adaptation Training Variable Statistics
One of the first tasks in training a model in Exhaustive is the process of establishing statistics. Statistics are important for each variable as they are used in Z Score normalisation of variables, which is an important concept in Machine Learning generally. The statistics are however comprehensively recorded, which provides rich insight about the dataset and problem domain more generally.
Keep in mind that statistics are calculated after the initial sample, but before class filtering, which means that it the statistics are a representation of the domain at large, rather than the positive class.
Notice in the model performance tabs, Statistics:
Click on the Statistics tab to expose for each variable summary statistics and histograms:
The variables are presented in the order of their correlation between the class and variable descending (strongest first).
Note that the grid is hierarchical and each row may be expanded upon:
The expansion of the row is intended to present an analysis on the extent to which variables related to one another, so called multi-colinearity. Click to expand the row:
For each variable, the correlation to that variable is returned. The variables are ordered by absolute correlation, with strongest correlations first. Colour formatting exists on the following basis:
- Red > 0.9.
- Orange > 0.7.
- Else Green.
Understanding the relationships between variables gives opportunity to collapse some variables into each other, for example by updating Abstraction Rules.