Datasets:

---

xcodemind

/

webcode2m

like 32

Follow

xcodemind 4

Sorry, we plan to elaborate on the implementation of TreeBLEU in the new publication. We provide the code directly for your reference.
Here we use the calculated tree_rouge_1 as the TreeBLEU score (CodeBLEU also takes the recall score as metric exactly.).
All the code is detailed below:

class HTMLMulNode:
    def __init__(self, name):
        self.childs = []
        self.name = name
        self.parent = None
        self.depth = 0

    def add_child(self, ch):
        self.childs.append(ch)
        ch.parent = self
        ch.depth = self.depth + 1      

def html2tree(html: str, drop_leaves=True):
    soup = BeautifulSoup(html, "html.parser")
    nodes = []

    def dfs(html_element: Tag, parent: HTMLMulNode):
        name = html_element.name if html_element.name else str(html_element.strip())
        new_node = HTMLMulNode(name)
        nodes.append(new_node)
        if parent is None:
            new_node.depth = 0
        else:
            parent.add_child(new_node)
        if html_element.name and html_element.contents:
            for child in html_element.contents:
                if child and str(child).strip() and html_element is not child:
                    if not (drop_leaves and child.name is None):
                        dfs(child, new_node)
        return new_node

    if soup.html:
        dfs(soup.html, None)         
    return nodes

def dom_sim(ref:str, cand:str):
    """
    subtree matching [1],
    match_score = count(cand matched in ref)/count(subtrees number of ref)
    
    [1] Ren, Shuo et al. “CodeBLEU: a Method for Automatic Evaluation of Code Synthesis.” ArXiv abs/2009.10297 (2020): n. pag.
    """

    ref_tree_nodes = html2tree(ref)
    cand_tree_nodes = html2tree(cand)
    
    if len(ref_tree_nodes) == 0 or len(ref_tree_nodes) == 0:
        return 0,0
    
    def collect_all_subtrees(nodes, height=1):
        subtrees = []
        for node in nodes:
            if len(node.childs) == 0:
                continue
            names = [node.name.strip().lower()]
            for child in node.childs:
                names.append(child.name.strip().lower())            
            
            subtrees.append('_'.join(names))
        return subtrees
    
    ref_subtree_seqs = collect_all_subtrees(ref_tree_nodes)
    cand_subtree_seqs = collect_all_subtrees(cand_tree_nodes)
    
    match_count = 0
    for seq in set(cand_subtree_seqs):
        if seq in set(ref_subtree_seqs):
            match_count += 1
    
    tree_rouge_1 = match_count/len(set(ref_subtree_seqs))
    
    match_count = 0
    for seq in cand_subtree_seqs:
        if seq in set(ref_subtree_seqs):
            match_count += 1
        
    tree_bleu = match_count/len(cand_subtree_seqs)
    
    return tree_bleu, tree_rouge_1