In the examples like tensorflow_mnist or scikit_learn in advisor_client, the config file has the goal MINIMIZE. In the scripts the metric used is accuracy. Am I missing something? Shouldn't the goal be maximize?

See here: https://mlflow.org/docs/latest/models.html

We'd want to add these classes:

• mlflow.neuraxle.save_model()
• mlflow.neuraxle.log_model()
• mlflow.neuraxle.load_model()

In the documentation on sensitivity analysis on this page https://openmole.org/Sensitivity.html#Specificconstructor

there are two constructor examples, both set evaluation to model , however there is no definition / declaration of what is the model. I think it would be a Task, but the examples miss these Tasks.

At the moment, the lazy loading features of HParams are not documented. It'd be nice to have basic documentation on them.

Hey Seb, could you spend some time on writing more detailed contributing guide? Currently the guide is a bit general so a step-by-step guide on how to run experiments and see visualizations after those experiments would be great. I haven't touched randpot for quite some time so need to refresh (to see if some visualizations impro

Current save/load methods focus on dumping and loading the pipeline definition in its JSON form, but provide no means to save a fitted pipeline and load it later to make predictions, being the usage of pickle outside of the pipeline the only way to go.

Let's re-implement the save/load methods to save the whole pipeline instance, and move the current save functionality to a to_json method.

This is a R6 problem: abstract classes aren't well defined, and a user problem: users don't read documentation. However for example if I do the following:

ps <- ParamSet$new(list(ParamDbl$new("A")))
x <- Sampler$new(ps)
x$sample(2)

It isn't until the last step I get a confusing error message:

Error in private$.sample(n) : abstract

Can I suggest something like th