In this page, you can define and start the optimization experiments.
An optimization experiment is an experiment that involves a chosen
dataset and topic modeling algorithm whose hyperparameters (or a subset
of hyperparameters) are optimized with respect of a chosen evaluation metric.
Quick start:
- Select the folder in which you will save your experiments.
For each experiment you need to specify the name of the
experiment
and the name for the set of experiments (batch name). The
"batch" allows
you to organize experiments in the same cluster. This will be useful later for
visualization purposes.
- Select a Dataset among our preprocessed datasets. You can just
train a model, or you can train and test the model (we have already splitted
the datasets).
-
Select a Model among our predefined topic models. You can
choose
among a selection of classical topic models and neural topic models.
-
Select the hyperparameters of the model that you want to
optimize
and/or fix the value of the hyperpararameters that you like.
If you optimize a hyperparameter, you need to select the value
range
for the selected hyperparameter.
-
Select the Metric that you want to optimize (just one for now,
but we
plan to do multi-objective hyperparameter optimization soon) and its
corresponding
parameters.
-
Select the configuration of the parameters of Bayesian
Optimization.
Hint:
Click on names of the hyperparameters, the models or the metrics for obtaining
additional information.