KaneAI - World’s First E2E Software Testing Agent.Book a Demo Join TestMu Conference

How to use with_ref method in pyatom

Best Python code snippet using pyatom_python

compare_reward.py

Source:compare_reward.py

1import numpy as np2import os3from nltk import PorterStemmer, word_tokenize, sent_tokenize4from nltk.corpus import stopwords5from tqdm import tqdm6from scipy.stats import spearmanr, pearsonr, kendalltau7from pytorch_transformers import *8from sklearn.metrics.pairwise import cosine_similarity9from sentence_transformers import SentenceTransformer10import argparse1112from scorer.data_helper.json_reader import read_sorted_scores, read_articles, read_processed_scores, read_scores13from helpers.data_helpers import sent2stokens_wostop, sent2tokens_wostop, sent2stokens, text_normalization14from scorer.auto_metrics.metrics import bleu, meteor15from resources import RUNS_DIR, ROUGE_DIR, BASE_DIR, MODEL_WEIGHT_DIR16from scorer.auto_metrics.rouge.rouge import RougeScorer17from step1_encode_doc_summ import raw_bert_encoder18from rewarder import Rewarder1920def sts_bert_encoder(model, sent_list):21    if not isinstance(sent_list,list):22        assert isinstance(sent_list,str)23        sent_list = sent_tokenize(sent_list)24    vecs = model.encode(sent_list)25    return vecs262728def sts_bert_rewarder(model, text1, text2):29    vec_list1 = sts_bert_encoder(model,text1)30    vec_list2 = sts_bert_encoder(model,text2)31    avg_vec1 = np.mean(vec_list1,axis=0)32    avg_vec2 = np.mean(vec_list2,axis=0)33    return cosine_similarity(avg_vec1.reshape(1, -1), avg_vec2.reshape(1, -1))[0][0]343536def raw_bert_rewarder(model, tokenizer, text1, text2):37    v1 = raw_bert_encoder(model,tokenizer,[text1])38    v2 = raw_bert_encoder(model,tokenizer,[text2])39    return cosine_similarity(v1.reshape(1, -1), v2.reshape(1, -1))[0][0]404142def evaluate_metric(metric, stem, remove_stop, with_ref, prompt='overall'):43    ''' metrics that use reference summaries '''44    assert metric in ['ROUGE-1-F', 'ROUGE-1-R', 'ROUGE-2-F', 'ROUGE-2-R', 'ROUGE-L-F', 'ROUGE-L-R', 'ROUGE-SU*-F',45                      'ROUGE-SU*-R', 'bleu-1', 'bleu-2', 'bleu-3', 'bleu-4', 'bleu-5', 'meteor',46                      'infersent', 'bert-raw','bert-sts','bert-nli','bert-human','mover-1', 'mover-2', 'mover-smd']47    stemmed_str = "_stem" if stem else ""48    stop_str = "_removestop" if remove_stop else ""49    if with_ref:50        ranks_file_path = os.path.join('outputs', 'wref_{}{}{}_{}_rank_correlation.csv'.format(metric, stemmed_str, stop_str, prompt))51    else:52        ranks_file_path = os.path.join('outputs', 'woref_{}{}{}_{}_rank_correlation.csv'.format(metric, stemmed_str, stop_str, prompt))53    print('\n====={}=====\n'.format(ranks_file_path))5455    #if os.path.isfile(ranks_file_path):56        #return ranks_file_path5758    ranks_file = open(ranks_file_path, 'w')59    ranks_file.write('article,summ_id,human_score,metric_score\n')6061    sorted_scores = read_sorted_scores()62    input_articles, _ = read_articles()63    corr_data = np.zeros((len(sorted_scores), 3))6465    stopwords_list = set(stopwords.words("english"))66    stemmer = PorterStemmer()6768    if metric.startswith('infersent'):69        from scorer.auto_metrics.infersent_metric import InferSentScorer70        infers = InferSentScorer()71    elif metric.startswith('sent2vec'):72        from scorer.auto_metrics.sent2vec_metric import Sent2Vec73        s2v = Sent2Vec()74    elif metric.startswith('bert'):75        pass76        if 'human' in metric:77            rewarder = Rewarder(os.path.join(MODEL_WEIGHT_DIR,'sample.model'))78        elif 'sts' in metric:79            bert_model = SentenceTransformer('bert-large-nli-stsb-mean-tokens')80        elif 'nli' in metric:81            bert_model = SentenceTransformer('bert-large-nli-mean-tokens')82        else:83            #raw BERT84            bert_tokenizer = BertTokenizer.from_pretrained('bert-large-uncased')85            bert_model = BertModel.from_pretrained('bert-large-uncased')86    elif metric.startswith('mover'):87        print('Make sure that your have started the mover server. Find details at https://github.com/AIPHES/emnlp19-moverscore.')88        from summ_eval.client import EvalClient89        mover_scorer = EvalClient()9091    for i, (article_id, scores_list) in tqdm(enumerate(sorted_scores.items())):92        human_ranks = [s['scores'][prompt] for s in scores_list]93        if len(human_ranks) < 2: continue94        ref_summ = scores_list[0]['ref']95        article = [entry['article'] for entry in input_articles if entry['id']==article_id][0]9697        if stem and remove_stop:98            sys_summs = [" ".join(sent2stokens_wostop(s['sys_summ'], stemmer, stopwords_list, 'english', True)) for s in scores_list]99            ref_summ = " ".join(sent2stokens_wostop(ref_summ, stemmer, stopwords_list, 'english', True))100            article = " ".join(sent2stokens_wostop(article, stemmer, stopwords_list, 'english', True))101        elif not stem and remove_stop:102            sys_summs = [" ".join(sent2tokens_wostop(s['sys_summ'], stopwords_list, 'english', True)) for s in scores_list]103            ref_summ = " ".join(sent2tokens_wostop(ref_summ, stopwords_list, 'english', True))104            article = " ".join(sent2tokens_wostop(article, stopwords_list, 'english', True))105        elif not remove_stop and stem:106            sys_summs = [" ".join(sent2stokens(s['sys_summ'], stemmer, 'english', True)) for s in scores_list]107            ref_summ = " ".join(sent2stokens(ref_summ, stemmer, 'english', True))108            article = " ".join(sent2stokens(article, stemmer, 'english', True))109        else:110            sys_summs = [s['sys_summ'] for s in scores_list]111112        summ_ids = [s['summ_id'] for s in scores_list]113        sys_summs = [text_normalization(s) for s in sys_summs]114        ref_summ = text_normalization(ref_summ)115        article = text_normalization(article)116117        if 'rouge' in metric.lower():118            auto_metric_ranks = []119            for ss in sys_summs:120                rouge_scorer = RougeScorer(ROUGE_DIR,BASE_DIR)121                if with_ref: auto_metric_ranks.append(rouge_scorer(ss, ref_summ)[metric])122                else: auto_metric_ranks.append(rouge_scorer(ss, article)[metric])123        elif metric.startswith('bleu'):124            n = int(metric.split('-')[1])125            if with_ref: auto_metric_ranks = [bleu(ss, [ref_summ], n, smooth=False) for ss in sys_summs]126            else:  auto_metric_ranks = [bleu(ss, [article], n, smooth=False) for ss in sys_summs]127        elif metric.startswith('meteor'):128            if with_ref: auto_metric_ranks = [meteor(ss, [ref_summ]) for ss in sys_summs]129            else: auto_metric_ranks = [meteor(ss, [article]) for ss in sys_summs]130        elif metric.startswith('infersent'):131            if with_ref: auto_metric_ranks = [infers(ss, ref_summ) for ss in sys_summs]132            else: auto_metric_ranks = [infers(ss, article) for ss in sys_summs]133        elif metric.startswith('sent2vec'):134            if with_ref: auto_metric_ranks = [s2v.score(ss, ref_summ) for ss in sys_summs]135            else: auto_metric_ranks = [s2v.score(ss, article) for ss in sys_summs]136        elif metric.startswith('bert'):137            if 'human' in metric:138                if with_ref: auto_metric_ranks = [rewarder(ref_summ,ss) for ss in sys_summs]139                else: auto_metric_ranks = [rewarder(article,ss) for ss in sys_summs]140            elif 'sts' in metric or 'nli' in metric:141                if with_ref: auto_metric_ranks = [sts_bert_rewarder(bert_model,ss,ref_summ) for ss in sys_summs]142                else: auto_metric_ranks = [sts_bert_rewarder(bert_model,ss,article) for ss in sys_summs]143            else: #raw BERT encoder144                if with_ref: auto_metric_ranks = [raw_bert_rewarder(bert_model,bert_tokenizer,ss,ref_summ) for ss in sys_summs]145                else: auto_metric_ranks = [raw_bert_rewarder(bert_model,bert_tokenizer,ss,article) for ss in sys_summs]146        elif metric.startswith('mover'):147            if '1' in metric: mm = 'wmd_1'148            elif '2' in metric: mm = 'wmd_2'149            else: mm = 'smd'150            if with_ref: cases = [ [[ss], [ref_summ], mm] for ss in sys_summs ]151            else: cases = [ [[ss], sent_tokenize(article), mm] for ss in sys_summs ]152            auto_metric_ranks = mover_scorer.eval(cases)['0']153154        for sid, amr, hr in zip(summ_ids, auto_metric_ranks, human_ranks):155            ranks_file.write('{},{},{:.2f},{:.4f}\n'.format(article_id, sid, hr, amr))156157        spearmanr_result = spearmanr(human_ranks, auto_metric_ranks)158        print(spearmanr_result[0])159        pearsonr_result = pearsonr(human_ranks, auto_metric_ranks)160        kendalltau_result = kendalltau(human_ranks, auto_metric_ranks)161        corr_data[i, :] = [spearmanr_result[0], pearsonr_result[0], kendalltau_result[0]]162163    corr_mean_all = np.nanmean(corr_data, axis=0)164    print('\n====={}=====\n'.format(ranks_file_path))165    print("Correlation mean on all data spearman/pearsonr/kendall: {}".format(corr_mean_all))166167    ranks_file.flush()168    ranks_file.close()169170    return ranks_file_path171172def parse_args():173    ap = argparse.ArgumentParser("arguments for summary sampler")174    ap.add_argument('-m','--metric',type=str,default='mover-1',choices=['ROUGE-1-F', 'ROUGE-1-R', 'ROUGE-2-F', 'ROUGE-2-R', 'ROUGE-L-F', 'ROUGE-L-R', 'ROUGE-SU*-F',175                      'ROUGE-SU*-R', 'bleu-1', 'bleu-2', 'bleu-3', 'bleu-4', 'bleu-5', 'meteor',176                      'infersent', 'bert-raw','bert-sts','bert-nli','bert-human', 'mover-1', 'mover-2', 'mover-smd'],help='compare which metric against the human judgements')177    ap.add_argument('-p','--prompt',type=str,default='overall',help='which human ratings you want to use as ground truth',choices=['overall','grammar'])178    ap.add_argument('-r','--with_ref',type=int,default=0,help='whether to use references in your metric; 1: yes, 0: no')179    ap.add_argument('-s','--stem',type=int,help='whether stem the texts before computing the metrics; 1 yes, 0 no')180    ap.add_argument('-rs','--remove_stop',type=int,help='whether remove stop words in texts before computing the metrics; 1 yes, 0 no')181    args = ap.parse_args()182    return args.metric, args.prompt, args.with_ref, args.stem, args.remove_stop183184185if __name__ == '__main__':186    metric, prompt, with_ref, stem, remove_stop = parse_args()187    with_ref = bool(with_ref)188    stem = bool(stem)189    remove_stop = bool(remove_stop)190191    print('\n=====Arguments====')192    print('metric: '+metric)193    print('prompt: '+prompt)194    print('with ref: '+repr(with_ref))195    print('stem: '+repr(stem))196    print('remove stopwords: '+repr(remove_stop))197    print('=====Arguments====\n')198
...

score.py

Source:score.py

...62            if isinstance(self.src, Variable):63                with self.src.open_for_read(out) as srcref:64                    scaled, is_temp = self.dest.scale_other_to_this(self.src,65                                                                    srcref, out)66                    out.write(self.with_ref(ref, scaled))67                    if is_temp:68                        out.free_temp(scaled)69            else:70                self.apply_const_src(ref, self.dest.to_int(self.src), out)71            if not self.is_additive:72                self.dest.scale_down(ref, out)73    def apply_const_src(self, ref, val, out):74        if val < 0 and self.with_neg_const is not None:75            out.write(self.with_neg_const(ref, -val))76        else:77            out.write(self.with_const(ref, val))78    def run(self, ev):79        assert isinstance(self.dest, CompilerVariable)80        if isinstance(self.src, Variable):81            assert isinstance(self.src, CompilerVariable)82            src = self.src.get_value()83        else:84            src = self.src85        self.dest.set_value(self.constfunc(self.dest.get_value(), src))86    def serialize(self, holder):87        dest, src = self.serialize_args(holder)88        return '%s %s %s' % (dest, self.with_ref.op, src)89    __op_lookup = {}90    @classmethod91    def lookup_by_op(cls, op):92        if not len(cls.__op_lookup):93            for clz in get_subclasses(cls):94                if hasattr(clz, 'with_ref'):95                    cls.__op_lookup[clz.with_ref.op] = clz96        return cls.__op_lookup[op]97import operator98class OnlyRefOperationInsn(SimpleOperationInsn):99    def apply_const_src(self, ref, val, out):100        srcref = out.allocate_temp()101        out.write(c.SetConst(srcref, val))102        out.write(self.with_ref(ref, srcref))103        out.free_temp(srcref)104class AddScore(SimpleOperationInsn):105    with_ref = c.OpAdd106    with_const = c.AddConst107    with_neg_const = c.RemConst108    constfunc = operator.add109    identity = 0110    is_additive = True111class SubScore(SimpleOperationInsn):112    with_ref = c.OpSub113    with_const = c.RemConst114    with_neg_const = c.AddConst115    constfunc = operator.sub116    identity = 0...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.