[cyorir]

My guess is that eval and query have the worst performance, since they need to be interpreted first. As for the other methods, I'm not too sure how they rank. Personally, for style I agree with the first comment on the linked page, which is to split up the conditions for readability:

  f1 = (df[“col1”] == condition1)

  f2 = (df[“col2”] == condition2)

  df[f1 & f2]