Why was sponge_knight banned for a month? He did ask a series of rapid-fire questions that were downvoted. (I don't entirely agree with those, but I guess the questions might have annoyed someone, especially several in a row.) I also looked at his recent comments and didn't find anything objectionable (although offensive comments could have been deleted). Moreover, don't we usually suspend people for 24 hours or a week? A month seems severe.
@gung I think that's the same user that asked a bunch of relatively poor questions a few months back -- he tends to get very defensive and hostile when people are critical. So that's probably what happened. He's been suspended for shorter terms at least once before.
There is some discussion in the comments here stats.stackexchange.com/a/226395/28666 about whether smaller p-values are "move convincing" (an old topic). Perhaps somebody will be interested.
Interested to get people's thoughts on the following scenario:
I have a bunch of text data and want to use tfidf vectorizer to create some features and then run a model. Should I run the vectorizer and create the features then split into a train and test set or do I have to split before making the features - running tfidf independently on the train and test sets?
@user5552894 You have to split first and create your features using only the train set, otherwise the structure of your test set will be, at least somewhat, known to your learning algorithm.
Im not sure what tfidf is though, but the general principle should apply.