Regularisation framework containers.
Object of this class are created with the respective regularisation
function: knnOptimisation
,
svmOptimisation
, plsdaOptimisation
,
knntlOptimisation
, ...
algorithm
:Object of class "character"
storing
the machine learning algorithm name.
hyperparameters
:Object of class "list"
with
the respective algorithm hyper-parameters tested.
design
:Object of class "numeric"
describing
the cross-validation design, the test data size and the number of
replications.
log
:Object of class "list"
with warnings
thrown during the hyper-parameters regularisation.
seed
:Object of class "integer"
with the random
number generation seed.
results
:Object of class "matrix"
of
dimenstions times
(see design
) by number of
hyperparameters + 1 storing the macro F1 values for the respective
best hyper-parameters for each replication.
f1Matrices
:Object of class "list"
with
respective times
cross-validation F1 matrices.
cmMatrices
:Object of class "list"
with
respective times
contingency matrices.
testPartitions
:Object of class "list"
with
respective times
test partitions.
datasize
:Object of class "list"
with details
about the respective inner and outter training and testing data
sizes.
Only in ThetaRegRes
:
predictions
: A list
of predictions for the
optimisation iterations.
otherWeights
: Alternative best theta weigts: a vector
per iterations, NULL
if no other best weights were found.
Returns a matrix of F1 scores for the optimisation parameters.
signature(object = "GenRegRes", t =
"numeric")
and signature(object = "ThetaRegRes", t =
"numeric")
: Constructs a table of all possible parameter
combination and count how many have an F1 scores greater or equal
than t
. When t
is missing (default), the best F1
score is used. This method is useful in conjunctin with
plot
.
Returns the best parameters. It is however strongly
recommended to inspect the optimisation results. For a
ThetaRegRes
optimisation result, the method to chose the
best parameters can be "median"
(default) or "mean"
(the median or mean of the best weights is chosen), "max"
(the first weights with the highest macro-F1 score, considering
that multiple max scoring combinations are possible) or
"count"
(the observed weight that get the maximum number of
observations, see f1Count
). The favourP
argument can
be used to prioritise weights that favour the primary data
(i.e. heigh weights). See favourPrimary
below.
Returns the seed used for the optimisation run.
signature(object = "GenRegRes")
: Returns a
vector of recorded warnings.
signature(object = "GenRegRes")
: Plots a
heatmap of of the optimisation results. Only for
"GenRegRes"
instances.
Plots the optisisation results.
Shows the object.
Only for ThetaRegRes
:
combineThetaRegRes(object)
Takes a list
of
ThetaRegRes
instances to be combined and returnes a new
ThetaRegRes
instance.
favourPrimary(primary, auxiliary, object, verbose =
TRUE)
Takes the primary
and auxiliary
data
sources (two MSnSet
instances) and a
ThetaRegRes
object and returns and updated
ThetaRegRes
instance containing best parameters/weigths
(see the getParams
function) favouring the primary data
when multiple best theta weights are available.
Laurent Gatto <lg390@cam.ac.uk>
1 2 |
Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.
Please suggest features or report bugs with the GitHub issue tracker.
All documentation is copyright its authors; we didn't write any of that.