Use pandas Magic for the FrequencyExperimenter

[crantila]

What we have works, but it would probably be faster to use the Series.value_counts() method as described here.

[crantila]

@alexandermorgan Do you think this is worth exploring?

[alexandermorgan]

I think so. In any event evaluating whether or not this is worth more attention shouldn't take long at all. I'll try to get to this tomorrow.

[alexandermorgan]

Though it could sped up, since what we have works I don't think this is high priority. If we ever get around to doing a rewrite of the ngram indexer we could maybe do this one in the same sitting. Otherwise let's just keep what we've got.

[alexandermorgan]

Note that the new_ngram indexer puts all voice-combination analyses done on the same piece in different columns in the same dataframe, rather than in separate dataframes. This will hopefully simplify the counting of ngrams. Also, when I want to count events quickly in a multi-indexed dataframe, I typically use this command:

df.stack().stack().value_counts()

The two stack() commands strip off successive layers of our vis multi-indices, leaving you with a series at the end. If we use this line of code at any point in the future, we may want to make it a recursive call, i.e. something like this:

def counts(df):
    if type(df) == pandas.DataFrame:
        df = df.stack()
        return counts(df)
    else:
        return df.value_counts()