Online Retraining

Select a model.
Respiratory Events
Select a type of scoring.
Select a previously trained model.
Train a new predictive model for Sleep Staging, Arousal, or Respiratory Events scoring of polysomnographic recording. While training is in progress, you may navigate away from the page or close your browser. Trained models may be saved for later use. Select an existing model to test on out-of-sample data, or configure and train a new model.
Model configuration.
Configuration
Multiclass models use the softmax function to produce calibrated probabilities for each Sleep Stage (Stage Wake, N1, N2, N3, or REM), Arousal, or Respiratory Event (Central and Obstructive Apnea or Hypopnea), for each analysed epoch (Sleep Staging) or window (Arousals and Respiratory Events) of recording. More hidden layers can learn more abstract, hierarchical patterns but increase training time and overfitting. Fewer layers are faster and simpler, but may underfit complex data. Hidden units control layer capacity. More units can learn more complex patterns but increase training time and overfitting. Fewer units reduce training time and produce a simpler model. Activation functions control how each layer transforms signals and introduce non-linearity, allowing the model to learn complex patterns. Activations can affect training time, stability, and accuracy.
Class imbalance (highly uneven class representation within the training data) may be corrected by adjusting class weights.
Model Information
Feature Selection
Feature Description
Phase Mean Difference Difference between the Phase Mean of the current window and a 40-second baseline.
Phase Mean Mean of the Phase.
Phase Positive Fraction Fraction of the Phase above a positive deadband.
Phase Negative Fraction Fraction of the Phase below a negative deadband.
Phase Positive Area Area of the Phase above a positive deadband.
Phase Negative Area Area of the Phase below a negative deadband.
Phase Spread Difference Difference between the Phase Spread of the current window and a 40-second baseline.
Phase RMS Difference Difference between the Phase RMS of the current window and a 40-second baseline.
Phase Spread Spread (P90 - P10) of the Phase.
Phase RMS Root mean square of the Phase.
Phase Permutation Entropy Permutation entropy of the Phase where the embedding delay equals 1, and m equals 3. A dynamic measure of regularity.
Sum Inspiratory Area Difference Difference between the Sum Inspiratory Area of the current window and a 40-second baseline.
Sum Inspiratory Area Inspiratory Area of the Sum.
Sum SSFD Sum of squared first differences in the Sum.
Sum Zero Crossings Number of sign changes between consecutive samples of the Sum.
Sum Spread Difference Difference between the Sum Spread of the current window and a 40-second baseline.
Sum RMS Difference Difference between the Sum RMS of the current window and a 40-second baseline.
Sum Spread Spread (P90 - P10) of the Sum.
Sum RMS Root mean square of the Sum.
Sum Permutation Entropy Permutation entropy of the Sum where the embedding delay equals 1, and m equals 3. A dynamic measure of regularity.
Flow Inspiratory Peak Difference Difference between the Inspiratory P95 of the Flow in the current window and a 40-second baseline.
Flow Inspiratory Area Difference Difference between the Inspiratory Area of the Flow in the current window and a 40-second baseline.
Flow Inspiratory Peak Inspiratory P95 of the Flow.
Flow Inspiratory Area Inspiratory Area of the Flow.
Flow Inspiratory Ratio Ratio of the mean Inspiratory Flow to the Inspiratory P95.
Flow Inspiratory Fraction Fraction of Inspiratory Flow above 80% of the Inspiratory P95.
Flow SSFD Sum of squared first differences in the Flow.
Flow Zero Crossings Number of sign changes between consecutive samples of the Flow.
Flow Spread Difference Difference between the Flow Spread (P90 - P10) of the current window and a 40-second baseline.
Flow RMS Difference Difference between the Flow RMS of the current window and a 40-second baseline.
Flow Spread Spread (P90 - P10) of the Flow.
Flow RMS Root mean square of the Flow.
Flow Permutation Entropy Permutation entropy of the Flow where the embedding delay equals 1, and m equals 3. A dynamic measure of regularity.
Oronasal Inspiratory Peak Difference Difference between the Inspiratory P95 of the Oronasal Flow in the current window and a 40-second baseline.
Oronasal Inspiratory Area Difference Difference between the Inspiratory Area of the Oronasal Flow in the current window and a 40-second baseline.
Oronasal Inspiratory Peak Inspiratory P95 of the Oronasal Flow.
Oronasal Inspiratory Area Inspiratory Area of the Oronasal Flow.
Oronasal SSFD Sum of squared first differences in the Oronasal Flow.
Oronasal Zero Crossings Number of sign changes between consecutive samples of the Oronasal Flow.
Oronasal Spread Difference Difference between the Oronasal Flow Spread (P90 - P10) of the current window and a 40-second baseline.
Oronasal RMS Difference Difference between the Oronasal Flow RMS of the current window and a 40-second baseline.
Oronasal Spread Spread (P90 - P10) of the Oronasal Flow.
Oronasal RMS Root mean square of the Oronasal Flow.
Oronasal Permutation Entropy Permutation entropy of the Oronasal Flow where the embedding delay equals 1, and m equals 3. A dynamic measure of regularity.
Select training data.
Select a precomputed dataset to use for model training. Only datasets specifically configured for training are available. Cross-validation results are available during model testing.
Training configuration and summary.
Bayesian Optimization
Number of objective function evaluations.
Number of candidates evaluated per iteration.
Configure the Bayesian Optimizer and search space for each hyperparameter. The number of iterations determines how many points are evaluated, and the number of candidates controls the resolution of the acquisition search.
Training Progress
Training in progress. You may use the analysis functionality or close your browser.

Optimization Iteration 5 of 20

Current Parameters: Lambda=0
0
0.4
0.8
Iteration
Multiclass Log Loss
Bayesian Optimization is used to search the hyperparameter ranges provided. This process attempts to balance exploration and exploitation to select the optimal values for the trained model. Gaussian Process Regression (GPR) and the Expected Improvement acquisition function are used. For each set of hyperparameters, models are trained using the appropriate solver. K-fold cross-validation is used to produce a validation error which is input into the GPR that underpins the Bayesian Optimization. The most optimal hyperparameters, as determined by the optimization process, are then used to train the final model. You may save the trained model when training is complete. Saved models are accessible immediately through the Analysis area of the platform.