ODYSSEE Quasar – Solver
- Added a new library dll containing the following functions for Neural Networks methodologies:
- MLP: Multi-Layer Perceptron model proposes the training and the prediction steps.
- predictMLP: Multi-Layer Perceptron model proposes only the prediction step (it supposes the
weight are already existing). - LSTM: Long Short Term Memory (Recurrent Neural Network) model proposes the training
and the prediction steps for the time-dependent data. - predictLSTM: Long Short Term Memory (Recurrent Neural Network) model proposes only the
prediction step
- Added a new function gpnewpoints in Quasar External, to choose new points for the DOE among specified candidates thanks to a Gaussian approach.
- Added new function newpoints in Quasar External, which proposes new points following a gradient approach.
- Added a new function canny in Quasar Image. It applies Canny filter, to extract the contours of the image.
- Added new useful functions in the qsr library:
- isInsideMatrix(double a, Matrix b) to find if a value is inside a matrix
- calculateEuclDistance(Matrix A, Matrix B) to calculate an Euclidian distance
- bubbleSort(Matrix A, int orientation, double direction, int idRef) to sort a matrix with the bubble algorithm
- argSort(Matrix A, int orientation, double direction, int idRef) to return an Id list corresponding to the order of the values in the selected column or row
- range1D(int n, int axis) to create a 1d matrix with values from 0 to n the dimension asked.
- removeLine(Matrix A, int j) to remove the line j from the matrix A
- The Quasar documentation has been updated with the following functions:
- Cstdxm resizes the curve in specified number of points
- Inrange checks if XN matrix is in the range of X
- Sort_by_rows sorts the data by column ID
- Order_by_row_index_X sorts the Y file by the column ID from X
- RegressionQR calculates the regression coefficients when the matrix is ill-conditioned
(often due to dependencies between parameters) - Condition number calculates the conditional number of the matrix (a high conditional number
means the matrix is ill-conditioned) - BellTransform converts a curve in indicators specific to curves with peaks
- BellTransformReverse from specific indicators, rebuilds a curve with peaks
- For the mining externals for correlation, covariance, and dependencies, when the input matrices do not have the same number of rows, a warning message is displayed.
- Bug Fixes
- For the Inverse Distance algorithm, an update has been done for linear interpolation with 1 Neighbor (INVD1, Neighbor = 1).
- For matrices over 600 lines, in certain condition, with a pod execution, it could have NAN results. An evolution on it has been done.