Created by: jeff41404

程序如下，是1个简单的线性回归模型

import paddle
import paddle.fluid as fluid
import numpy as np

# init paddle
place = fluid.CPUPlace()

def inference_program():
    # input data
    x = fluid.layers.data(name='x', shape=[1], dtype='float32')
    # define network
    y_predict = fluid.layers.fc(input=x, size=1, act=None, bias_attr=True)
    return y_predict

def train_program():
    y_predict = inference_program()
    # label
    y = fluid.layers.data(name='y', shape=[1], dtype='float32')
    # cost function
    cost = fluid.layers.square_error_cost(input=y_predict, label=y)
    avg_cost = fluid.layers.mean(cost)
    #return avg_cost
    accuracy = fluid.layers.accuracy(input=y_predict, label=y)
    return [avg_cost, accuracy]

def optimizer_program():
    # create optimizer
    optimizer = fluid.optimizer.Momentum(momentum=0.6, learning_rate=0.001/64)
    return optimizer

# create trainer
trainer = fluid.Trainer(
        train_func=train_program,
        optimizer=optimizer_program(),
        place=place)

# event_handler to print training info
def event_handler(event):
    if isinstance(event, fluid.EndStepEvent):
        if event.step % 1 == 0:
            print "Pass %d, Batch %d, Cost %f" % (
                event.epoch, event.step, event.metrics[0])
            #print "Pass %d, Batch %d, Cost %f, Acc %f" % (
                #event.epoch, event.step, event.metrics[0], event.metrics[1])
    # product model every 10 pass
    if isinstance(event, fluid.EndEpochEvent):
        if event.epoch % 100 == 0:
            trainer.save_params('/exam/fluid_model/params_pass_%d.tar' % (event.epoch))


# define training dataset reader
def train_reader():
    train_x = []
    train_y = []
    #train_x = np.array([[1], [2], [3], [5]])
    #train_y = np.array([[6], [8], [10], [14]])

    with open('/exam/train.data', 'r') as f:
        lines=f.readlines()
        for line in lines:
            lineData=line.strip().split(',')
            train_x.append(lineData[0])
            train_y.append(lineData[1])
        train_x=np.array(train_x).astype("float32")
        train_y=np.array(train_y).astype("float32")

    def reader():
        for i in xrange(train_y.shape[0]):
            yield train_x[i], train_y[i]

    return reader


# define feeding map
feeding = ['x', 'y']

# training
trainer.train(
    reader=paddle.batch(
        train_reader(), batch_size=10),
    feed_order=feeding,
    event_handler=event_handler,
    num_epochs=2001)

# inference
# loading the model which generated by training
model_dir = 'fluid_model/params_pass_2000.tar'

# Input multiple sets of data, Output the infer result in a array.
x = [ [-3], [13], [6], [7], [20] ]
print ("inference data : %s , expecting print : -2, 30, 16, 18, 44" % x)
infer_x = np.array(x).astype("float32")
inferencer = fluid.Inferencer(infer_func=inference_program, param_path=model_dir, place=place)
print "inference result :"
print inferencer.infer(inputs={'x': infer_x})

============================================= 运行时报错： 主要问题是： paddle.fluid.core.EnforceNotMet: Tensor holds the wrong type, it holds f at [/paddle/paddle/fluid/framework/tensor_impl.h:44] 尝试在如上程序中改动了1点，发现能绕过去 即 train_program 中 return [avg_cost, accuracy] 改为 return avg_cost ，运行程序就不报错了 请教一下是什么问题