src/test/scripts/functions/builtin/HyperbandLM.dml

#-------------------------------------------------------------
#
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.
#
#-------------------------------------------------------------

l2norm = function(Matrix[Double] X, Matrix[Double] y, Matrix[Double] B) return (Matrix[Double] loss) {
 loss = as.matrix(sum((y - X%*%B)^2));
}

X = read($1);
y = read($2);

# size of dataset chosen such that number of maximum iterations influences the 
# performance of candidates
numTrSamples = 100;
numValSamples = 100;

X_train = X[1:numTrSamples,];
y_train = y[1:numTrSamples,];
X_val = X[(numTrSamples+1):(numTrSamples+numValSamples+1),];
y_val = y[(numTrSamples+1):(numTrSamples+numValSamples+1),];
X_test = X[(numTrSamples+numValSamples+2):nrow(X),];
y_test = y[(numTrSamples+numValSamples+2):nrow(X),];

params = list("reg", "tol");

# only works with continuous hyper parameters in this implementation
paramRanges = matrix(0, rows=2, cols=2);

paramRanges[1,1] = 0;
paramRanges[1,2] = 20;
paramRanges[2,1] = 10^-10;
paramRanges[2,2] = 10^-12;

# use lmCG, because this implementation of hyperband only makes sense with
# iterative algorithms
[B1, optHyperParams] = hyperband(X_train=X_train, y_train=y_train, X_val=X_val,
  y_val=y_val, params=params, paramRanges=paramRanges, R=50, eta=3, verbose=TRUE);

# train reference with default values
B2 = lmCG(X=X_train, y=y_train, verbose=FALSE);

l1 = l2norm(X_test, y_test, B1);
l2 = l2norm(X_test, y_test, B2);
R = as.scalar(l1 <= l2);

write(R, $3)