Each version of Monument adds capabilities and robustness to the platform.
Upgrade at no cost to the latest version by following the link at Help > Check for Updates...
Q1 2021 - User-defined Columns and Stacked Models(release date 2021-03-05)
- User-defined columns with user-defined functions
- Over 40 functions to modify and enhance your data, listed here at User-defined Columns
- Modify your training data within Monument
- Create additional data features to enhance model performance
- Model stacking
- Capabilities to stack, combine, ensemble, and weight models
- Estimate cluster groupings
- Create estimates of error rates
- Code signing
- Increased security across Windows and macOS platforms via code-signing
Q4 2020 - Model Serving(release date 2020-12-30)
- Model serving
- Run your trained model on new, out-of-sample data
Q4 2020 - Autopilot and Model Comparison(release date 2020-12-08)
- Autopilot: automatic algo selection
- Select your target variable and let Monument analyze which analytic methods are most suitable
- Model benchmarking and comparison
- Is this model useful? Determine the performance of your model relative to a naive benchmark
- Is this model better than the prior edition
- Multiple plots and nested plots
- Multiple dimensions now show multiple plots or nested data, as appropriate in the UI
Q4 2020 - LSTM Variants(release date 2020-11-16)
- Multi-layer Deep LSTMs
- LSTMs with multiple hidden LSTM layers where each layer contains multiple memory cells, each fed into the subsequent layer
- Phased LSTM
- LSTM variation emulating the spiking nature of neurons belonging to 3rd Gen of neural nets
- **Rescaled LSTM*
- LSTM variation that enhances LSTM memory performance with Gaussian-Process downsampling
Q3 2020 - Cross-sectional Data Handling(release date 2020-08-25)
- Cross-sectional and panel data algorithms for both numerical and categorical data
- Adaptive algo display with predictive methods that automatically adapt to the currently selected data (for example, displaying a date enables timeseries methods)
- Algos are shown only where appropriate for the selected data.
- When a Dimension is moved into the charting area, only classification algos are shown and run.
- When a time pill is moved into COLS, only timeseries algos are shown.
- New cross-sectional data methods
- LGBM for cross-sectional and panel data, with AutoML
- SVM for categorical data and SVR for numerical cross-sectional and panel data
- Feature importance for LGBM, G-Boost, and Ent-Boost
- Sample datasets available at Help > Demos
- Further training and guidance available at Help > FAQ and YouTube links
- Error edge-case cleanup with 95% reduction in error messages, supported by enhanced, user-driven error reporting
Q3 2020 - Windowing
(release date 2020-07-20)
- Partitioning methods in windowing enable out-of-sample cross validation across an entire time series
- Shuffle windowing methodology cleanup with shuffle discontinued in dlm, lightbml2, tvar, gsboost, entboost. Shuffle in these instances enabled look-forward bias.
- Binance-third-party data integration
- Binance rewritten to allow any valid starting date (previously limited to a single call, typically 500 entries). Also checks for existing values to minimize re-reads.
- Bug fixes