def log_iter()

in utils.py [0:0]

• 32 lines of code
• 4 McCabe index (conditional complexity)

    def log_iter(self, iter_no, nb_batches_per_iter, loss_train, loss_val,
                 elapsed, model):
        step = (iter_no + 1) * nb_batches_per_iter
        self._log(title='step', value=step)
        msg = 'steps: {}'.format(step)
        if self.data_unit == 'bpc':
            train_bpc = float(loss_train / math.log(2))
            val_bpc = float(loss_val / math.log(2))
            msg += '\ttrain: {:.3f}bpc\tval: {:.3f}bpc'.format(train_bpc, val_bpc)
            self._log(title='train_bpc', value=train_bpc)
            self._log(title='val_bpc', value=val_bpc)
        else:
            train_ppl = math.exp(loss_train)
            val_ppl = math.exp(loss_val)
            msg += '\ttrain: {:.2f}ppl\tval: {:.2f}ppl'.format(train_ppl, val_ppl)
            self._log(title='train_ppl', value=train_ppl)
            self._log(title='val_ppl', value=val_ppl)
        msg += '\tms/batch: {:.1f}'.format(elapsed)

        if model.module.layers[0].attn.attn.adapt_span_enabled:
            avg_spans = []
            max_spans = []
            for layer in model.module.layers:
                avg_spans.append(
                    layer.attn.attn.adaptive_span.get_current_avg_span())
                max_spans.append(
                    layer.attn.attn.adaptive_span.get_current_max_span())
            span_avg = float(sum(avg_spans)) / len(avg_spans)
            span_max = float(max(max_spans))
            self._log('span_avg', span_avg)
            self._log('span_max', span_max)
            msg += "\tspan_avg: {:.0f}\tspan_max: {:.0f}".format(span_avg, span_max)

        print(msg)