MixtureMapping

Train Gaussian Mixture Mappings

Provides:

1. Layers to build tensorflow models to map Gaussian mixtures
2. Tools to compute yield values of Gaussian mixtures in complex binning schemes

Example

import mixturemapping as mm  
import tensorflow as tf

inMeans = tf.keras.Input(shape=(mixN, inputMixM), name="Means", dtype=dataType)
inStdDevs = tf.keras.Input(shape=(mixN, inputMixM), name="StdDevs", dtype=dataType)
inWeight = tf.keras.Input(shape=(mixN), name="Weights", dtype=dataType)

mapModel = tf.keras.Sequential()
mapModel.add( tf.keras.layers.Dense(40, activation="relu", kernel_regularizer=regularizers.l2(0.001)) )
mapModel.add( tf.keras.layers.Dense(40, activation="relu", kernel_regularizer=regularizers.l2(0.001)) )
mapModel.add( tf.keras.layers.Dense(outputMixM))
y = mapModel(inMeans)

deltaModel = tf.keras.Sequential()
deltaModel.add( tf.keras.layers.Dense(40, activation="relu", kernel_regularizer=regularizers.l2(0.001)) )
deltaModel.add( tf.keras.layers.Dense(40, activation="relu", kernel_regularizer=regularizers.l2(0.001)) )
deltaModel.add( tf.keras.layers.Dense(outputMixM))
yDelta = deltaModel(inMeans)

covALayer = mm.layers.TrainableCovMatrix(outputMixM, name="CovA")
covA = covALayer(inMeans)

mapLayer = mm.layers.GeneralMapping(outputMixM, name="Mapping", dtype=dataType)
newDist = mapLayer({'means': inMeans, 'y':y, 'yDelta':yDelta, 'stdDevs': inStdDevs, 'weights': inWeight, 'covA': covA})

distLayer = mm.layers.Distribution(dtype=dataType, regularize_cov_epsilon=0.95)
dist = distLayer(newDist)

Developement

$ py -m venv env
$ .\env\Scripts\activate
$ pip install -r requirements.txt