当数据量小于 batch_size 的时候，会出错

按照 README 中的脚本去进行 eval 的时候会出现
```
Traceback (most recent call last):
File "sentiment_classify.py", line 288, in
main(args)
File "sentiment_classify.py", line 273, in main
args.model_path)
File "sentiment_classify.py", line 210, in eval_net
class2_acc = class2_acc / (total_count - neu_count)
ZeroDivisionError: float division by zero
```

经过分析发现是自带数据集中测试数据集只有 200 条， 而默认 batch_size 是 256, 而且因为代码使用的 batch 是 paddle.v2.batch 而不是 paddle.batch， （这个两个函数的实现一样，但是默认参数在 paddle.v2.batch 里面 drop_last 默认是 True, paddle.batch 里默认是 False）， 所以会导致 test_reader 不会 yield.

建议更新代码，把 `utils.py` 中的

`import paddle.v2 as paddle`
改成 `import paddle`

或者在

`utils.py`101 和105 行中把
```
101 train_reader = paddle.batch(data_reader(data_path, word_dict, True), batch_size)
105 test_reader = paddle.batch(data_reader(data_path, word_dict, False), batch_size)
```
中加上 drop_last=False.

Start with utils.py lines 101 and 105 and reproduce the README eval using the bundled dataset and its default batch size. Check that the test reader yields the 200-item dataset instead of dropping it, and confirm eval_net no longer reaches a zero-denominator accuracy calculation.

Written by the indexing model from the issue text.