Diego Kobylkin

Wasn't it the image map that @ZXoho meant? It is mentioned here https://www.wordclouds.com/faq/

This should be fixed upstream already - need to test. There is no immediate impact on the current code as it works around this bug in a transparent way.

The upstream issue is fixed, but the workaround code is still in the project. We will have to remove it at some point. The workaround code is only invoked if...

You have a point on interpolation meaning. However the model are combined into ensembles so there might be some meaning to that too. Yet visually I like the line plot...

The essence of the service for the Receiver is to receive a file from the Sender and save it on his local file system. It seems logical to assume saving...

> A general heuristic you coud try is to run some basic scikit-learn models and set the timeout to be twice their runtime This can be very handy, actually. Great...

Cool! this will do for per_run_time_limit. Is there a programmatic way to get to the resulting loss function values from the various AutoSklearnClassifier seeds and their run number? What I...

Here is the heuristic that gets >0.98 performance for the parallel_example.py in about 400 seconds ``` import autosklearn.pipeline.components.classification clfs=autosklearn.pipeline.components.classification._classifiers from time import time max_clf_time=1 # second for clf_name,clf_class in clfs.items():...

I have added the heuristic to determine the per_run_time_limit based on the actual run time for classifiers for a given dataset here https://github.com/Motorrat/auto-sklearn/blob/patch-1/example/redly.py