Feature Description

We want to enable the users to specify the value ranges for any argument in the blocks.
The following code example shows a typical use case.
The users can specify the number of units in a DenseBlock to be either 10 or 20.

Code Example

import au

From issue #1302, it appears autosklearn is a bit unstable when run many times in the same script, i.e. in a for loop.

for i in range(400):
    automodel = AutoSklearn(full_resources)
    automodel.fit(x, y)

We currently have no test for this and it would be good to see if we can reproduce the same connection refused error.

• With Featuretools 1.0.0 we add a dataframe to an EntitySet with the following:

es = ft.EntitySet('new_es')

es.add_dataframe(dataframe=orders_df,
                 dataframe_name='orders',
                 index='order_id',
                 time_index='order_date')

Improvement

• However, you could also change the EntitySet setter to add it with this approach:

es = ft.Ent

When running TabularPredictor.fit(), I encounter a BrokenPipeError for some reason.
What is causing this?
Could it be due to OOM error?

Fitting model: XGBoost ...
-34.1179 = Validation root_mean_squared_error score
10.58s = Training runtime
0.03s = Validation runtime
Fitting model: NeuralNetMXNet ...
-34.2849 = Validation root_mean_squared_error score
43.63s =

Problem
Some of our transformers & estimators are not thoroughly tested or not tested at all.

Solution
Use OpTransformerSpec and OpEstimatorSpec base test specs to provide tests for all existing transformers & estimators.

There are several evaluation metrics that would be particularly beneficial for (binary) imbalanced classification problems and would be greatly appreciated additions. In terms of prioritizing implementation (and likely ease of implementation I will rank-order these):

1. AUCPR - helpful in the event that class labels are needed and the positive class is of greater importance.
2. **F2 Scor

Python 3.10 has been released. We should test it. If all the dependencies support it, we should add it to CI.

The pipeline.fit/fit_from_scratch difference is quite confusing for new users.
So, common 'fit' should be preferred in all situations.
Cache-based fit should be moved to external function.