def run()

in senteval/probing.py [0:0]

• 36 lines of code
• 5 McCabe index (conditional complexity)

    def run(self, params, batcher):
        task_embed = {'train': {}, 'dev': {}, 'test': {}}
        bsize = params.batch_size
        logging.info('Computing embeddings for train/dev/test')
        for key in self.task_data:
            # Sort to reduce padding
            sorted_data = sorted(zip(self.task_data[key]['X'],
                                     self.task_data[key]['y']),
                                 key=lambda z: (len(z[0]), z[1]))
            self.task_data[key]['X'], self.task_data[key]['y'] = map(list, zip(*sorted_data))

            task_embed[key]['X'] = []
            for ii in range(0, len(self.task_data[key]['y']), bsize):
                batch = self.task_data[key]['X'][ii:ii + bsize]
                embeddings = batcher(params, batch)
                task_embed[key]['X'].append(embeddings)
            task_embed[key]['X'] = np.vstack(task_embed[key]['X'])
            task_embed[key]['y'] = np.array(self.task_data[key]['y'])
        logging.info('Computed embeddings')

        config_classifier = {'nclasses': self.nclasses, 'seed': self.seed,
                             'usepytorch': params.usepytorch,
                             'classifier': params.classifier}

        if self.task == "WordContent" and params.classifier['nhid'] > 0:
            config_classifier = copy.deepcopy(config_classifier)
            config_classifier['classifier']['nhid'] = 0
            print(params.classifier['nhid'])

        clf = SplitClassifier(X={'train': task_embed['train']['X'],
                                 'valid': task_embed['dev']['X'],
                                 'test': task_embed['test']['X']},
                              y={'train': task_embed['train']['y'],
                                 'valid': task_embed['dev']['y'],
                                 'test': task_embed['test']['y']},
                              config=config_classifier)

        devacc, testacc = clf.run()
        logging.debug('\nDev acc : %.1f Test acc : %.1f for %s classification\n' % (devacc, testacc, self.task.upper()))

        return {'devacc': devacc, 'acc': testacc,
                'ndev': len(task_embed['dev']['X']),
                'ntest': len(task_embed['test']['X'])}