NN is one of the AI techniques that can be used for instance to predict groups membership.
It is necessary to previously train the NN with known data. With XYOM, this machine learning (ML) step is accessible in the “MACHINE LEARNING” section.
Provided a “weights” file has been produced after the ML step, the group membership attribution is accessible (also through the “IDENTIFICATION” section).
1/ “MACHINE LEARNING”
You must enter 3 arguments
I. The (fake) unknown file (could be true unknown also, for identification)
II. The reference data file
III. The subdivision of the reference data
2/ “IDENTIFICATION” (cannot be performed without the previous ML step)
You must call 2 files
I. The unknown data file
II. The “Weights” file as produced by the ML step.
3/ I M P O R T A N T R E M A R K S for LANDMARKS DATA
The general instruction lines above do not directly apply when data are landmark-based shape variable, because shape variables derived from landmarks are sample dependent. Please see hereunder what steps you must follow to tentatively assign landmarks configurations to other such landmarks configurations.
MACHINE LEARNING of XYOM for LANDMARK-BASED SHAPE variables:
- Do not enter RAW COORDINATES of LANDMARKS or of PSEUDOLANDMARKS as inputs
- Unknown and reference matrices must have the same number of columns
- Landmarks must be in the same order for unknown and reference specimens.
- For landmark-based shape data, the identification of unknown should use exclusively the MACHINE LEARNING section of XYOM (it should NOT use the IDENTIFICATION section).
- Please consider the following steps (These steps are NOT automatically implemented by XYOM):
- 1. Concatenate raw landmarks of unknown and reference individuals
- 2. Perform a GPA (Generalised Procrustes Analysis) on these concatenated data (single file)
- 3. Split the file of Procrustes Residuals (ending by …_ORP.txt) into unknown and reference data (2 files)
- 4. Use these 2 files of Procrustes Residuals as input for MACHINE LEARNING process generating WEIGHTS, and see the report
- 5. Do not keep these weights as valid weights for new identification analyses
MACHINE LEARNING of XYOM for PSEUDOLANDMARK-BASED SHAPE data:
For pseudolandmark-based shape data (NEF), the process is easier because they are not sample dependent. The identification of unknown may use the MACHINE LEARNING section AND the IDENTIFICATION section.
Please consider the following steps for pseudolandmarks data (These steps are NOT automatically implemented by XYOM):
- 1. Compute the NEF from unknown data
- 2. Compute the NEF from reference data
- 3. Select the same number of columns for each file (unknown and reference).
- 4. MACHINE LEARNING section: compute and save the WEIGHTS
- 5. New unknown (NEF, same number of columns as learning input) could be tentatively identified using the IDENTIFICATION section of XYOM
See here -> the detailed MLP (multilayer perceptron ) configuration by XYOM (Please remember that to adjust the parameters of the multilayer perceptron in terms of number of hidden layers, number of neurons by layer, etc. you should test FAKE unknown to check for overfitting).
Copyright (c) 2015-2016 Ulf Biallas
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the “Software”), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE