Open a web page for manual validation and error correction of predicted abundances in samples.
correctError(zidb, classifier, data = zidbDatRead(zidb), mode = "validation",
fraction = 0.05, sample.min = 100, sample.max = 200, grp.min = 2,
random.sample = 0.1, algorithm = "rf", diff.max = 0.2, prop.bio = NULL,
reset = TRUE, result = NULL)addItemsToTrain(train, CtxSmp, add.mode = "SV+NSV", threshold = NA,
dropItemsToTrain = dropItemsToTrain)
dropItemsToTrain(train, cl, drop.nb)
activeLearning(train, add.mode = "SV+NSV", threshold = NA)
Path to a Zidb file.
A ZIClass object appropriate for this sample and the desired classification.
A ZIDat or a ZITest object matching that sample (by default, it is the ZIDat object contained in the zidb file).
The mode to use for error correction. By default, mode = "validation"
,
where particles are manually validated. mode = "demo"
is the same one, but it sorts particles
according to the Class variable in data, ignoring changes made in the user interface (so that one
can explain the logic of the process without care about how particles are manually resorted).
Finally, mode = "stat"
do not display the user interface at all and calculates all steps
directly to show gain from the process from 0 to 100% of the particles validated.
The fraction of items to validate at each step (1/20th by default).
Minimal number of items to take at each step.
Maximal number of items to take at each step. In case the sample contains a very large number of items, the number of particles that are validated at each step are constrained by this parameter, and consequently, the total number of steps becomes large than 1/fraction, but usually, error correction allows to stop earlier.
Minimal number of items to take for each group, on average.
Fraction of random sample considered, when validating suspect items.
Machine learning algorithm used to detect suspect items.
Maximum difference allowed between probabilities in first and second class before considering the item is suspect.
Weight to apply to the groups for considering them as suspects (use biological or external considerations to build this).
Do we reset analysis in the case a temporary file already exists for that sample (recommended).
Name of the object in the calling environment where the results will be stored (ZITest object).
If not provided or NULL
, it is the basename of the zidb file without extension plus
_valid
.
the training set to complete.
the contextual samples containing validated items.
the mode for adding items, "SV"
: Validated Suspects,
"NSV"
: Validated Non-Suspects, or "SV+NSV"
: both (by default)).
the maximal number of items in each class of training set. This is used to decide when to drop items for the reworked training set.
the function to use to drop items in the training set
(depending on threshold). By default, it is dropItemsToTrain()
, but it
could also be a custom function that has the same arguments.
the class to consider.
the number of items to drop.
correctError()
returns nothing. It is called for its side-effect to install a web interface
for manual validation of samples.