Created by: minhozhou

• 版本、环境信息：    1）PaddlePaddle版本：1.4.1    2）CPU    3）系统环境：centos -预测信息    1）使用fluid.io.load_persistables进行加载预测
• 问题描述：在infer的时候，metrics.Auc比layers.auc得出的auc值低10个百分点

1. auc预测代码： 在网络中的auc定义为 auc_var, batch_auc_var, auc_states = layers.auc(input=fc, label=self.datas[-1])

infer_program = fluid.default_main_program().clone(for_test=True)

    place = fluid.CPUPlace()
    exe = fluid.Executor(place)
    exe.run(fluid.default_startup_program())

    fluid.io.load_persistables(exe, args.model_output_dir + '/pass-', infer_program)
   inference_scope = fluid.Scope()

    def set_zero(var_name):
        param = inference_scope.var(var_name).get_tensor()
        param_array = np.zeros(param._get_dims()).astype("int64")
        param.set(param_array, place)

    for auc_state in auc_states:
        set_zero(auc_state.name)
    batch_id = 0
    py_reader.start()
    while True:
        try:
            results, auc_val = exe.run(infer_program, fetch_list=[cost, auc_var])
            if batch_id % 100 == 0:
                logger.info("TEST --> batch: {} loss:{} auc:{}".format(batch_id, results, auc_val))
        except fluid.core.EOFException:
            py_reader.reset()
            break

2.Auc预测代码

  infer_program = fluid.default_main_program().clone(for_test=True)

    place = fluid.CPUPlace()
    exe = fluid.Executor(place)
    exe.run(fluid.default_startup_program())

    fluid.io.load_persistables(exe, args.model_output_dir + '/pass-', infer_program)

    batch_id = 0
    py_reader.start()
    auc_metric = fluid.metrics.Auc("ROC")
    while True:
        try:
            results, y_pred, y_true = exe.run(infer_program, fetch_list=[cost, preds, label])
            auc_metric.update(y_pred, y_true)
            if batch_id % 100 == 0:
                logger.info("TEST --> batch: {} loss:{} auc:{}".format(batch_id, results, auc_metric.eval()))
        except fluid.core.EOFException:
            py_reader.reset()
            break