heatmap.py

'''
@author: Leila Arras
@maintainer: Leila Arras
@date: 21.06.2017
@version: 1.0+
@copyright: Copyright (c) 2017, Leila Arras, Gregoire Montavon, Klaus-Robert Mueller, Wojciech Samek
@license: see LICENSE file in repository root
'''

import matplotlib.pyplot as plt

def rescale_score_by_abs (score, max_score, min_score):
    """
    Normalize the relevance value (=score), accordingly to the extremal relevance values (max_score and min_score), 
    for visualization with a diverging colormap.
    i.e. rescale positive relevance to the range [0.5, 1.0], and negative relevance to the range [0.0, 0.5],
    using the highest absolute relevance for linear interpolation.
    """
    
    # CASE 1: positive AND negative scores occur --------------------
    if max_score>0 and min_score<0:
    
        if max_score >= abs(min_score):   # deepest color is positive
            if score>=0:
                return 0.5 + 0.5*(score/max_score)
            else:
                return 0.5 - 0.5*(abs(score)/max_score)

        else:                             # deepest color is negative
            if score>=0:
                return 0.5 + 0.5*(score/abs(min_score))
            else:
                return 0.5 - 0.5*(score/min_score)   
    
    # CASE 2: ONLY positive scores occur -----------------------------       
    elif max_score>0 and min_score>=0: 
        if max_score == min_score:
            return 1.0
        else:
            return 0.5 + 0.5*(score/max_score)
    
    # CASE 3: ONLY negative scores occur -----------------------------
    elif max_score<=0 and min_score<0: 
        if max_score == min_score:
            return 0.0
        else:
            return 0.5 - 0.5*(score/min_score)    
  
      
def getRGB (c_tuple):
    return "#%02x%02x%02x"%(int(c_tuple[0]*255), int(c_tuple[1]*255), int(c_tuple[2]*255))

     
def span_word (word, score, colormap):
    return "<span style=\"background-color:"+getRGB(colormap(score))+"\">"+word+"</span>"


def html_heatmap (words, scores, cmap_name="bwr"):
    """
    Return word-level heatmap in HTML format,
    with words being the list of words (as string),
    scores the corresponding list of word-level relevance values,
    and cmap_name the name of the matplotlib diverging colormap.
    """
    
    colormap  = plt.get_cmap(cmap_name)
     
    assert len(words)==len(scores)
    max_s     = max(scores)
    min_s     = min(scores)
    
    output_text = ""
    
    for idx, w in enumerate(words):
        score       = rescale_score_by_abs(scores[idx], max_s, min_s)
        output_text = output_text + span_word(w, score, colormap) + " "
    
    return output_text + "\n"