Created by: sandycs

今天进展到数字识别章节问题 我这么理解。对于softmax回归来说，输入实际上是一个28x28=784的一维矩阵 输出是一个10个元素的一维矩阵。 于是我对程序作了一些变化： 我的输入是随机的：

global dataX dataX = np.random.rand(1000,3) 我的输出是条件生成的： （逻辑很简单。就是写一些条件，当X满足某些条件的时候，Y是某个确定的值） global dataY dataY = np.zeros((1000,3)) intcount = 0 for e in dataX: if (e[0]>0.5 and e[1]>0.5 and e[2]>0.5): dataY[intcount] = [1.0,0,0] else: if (e[2] <0.5): dataY[intcount] = [0,1.0,0] else: dataY[intcount] = [0,0,1.0] intcount = intcount +1

然后我就把这些输入进去进行训练了。

global UCI_TRAIN_DATA 
UCI_TRAIN_DATA = dataFinal[:800]
global UCI_TEST_DATA 
UCI_TEST_DATA= dataFinal[800:]
    
x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(3))
y = paddle.layer.data(name='y', type=paddle.data_type.dense_vector(3))

predict = softmax_regression(x)

cost = paddle.layer.classification_cost(input=predict, label=y)

# Save the inference topology to protobuf.
inference_topology = paddle.topology.Topology(layers=predict)
with open("inference_topology.pkl", 'wb') as f:
    inference_topology.serialize_for_inference(f)

parameters = paddle.parameters.create(cost)    
# create optimizer
# optimizer = paddle.optimizer.Momentum(momentum=0)
optimizer = paddle.optimizer.Momentum(
    learning_rate=0.1 / 3.0,
    momentum=0.9,
    regularization=paddle.optimizer.L2Regularization(rate=0.0005 * 3))

trainer = paddle.trainer.SGD(
    cost=cost, parameters=parameters, update_equation=optimizer)
# feeding = {'x': 0, 'y': 1}
# event_handler to print training and testing info
def event_handler(event):
    if isinstance(event, paddle.event.EndIteration):
        if event.batch_id % 100 == 0:
            print "Pass %d, Batch %d, Cost %f, %s" % (event.pass_id, event.batch_id, event.cost, event.metrics)
    if isinstance(event, paddle.event.EndPass):
        if event.pass_id % 10 == 0:
            with open('params_pass_%d.tar' % event.pass_id, 'w') as f:
                trainer.save_parameter_to_tar(f)
        result = trainer.test(
            reader=paddle.batch(
            testdata(), batch_size=2)
            )
        print "Test with Pass %d, Cost %f, %s\n" % (event.pass_id, result.cost, result.metrics)

# training
trainer.train(
    reader=paddle.batch(
        paddle.reader.shuffle(traindata(), buf_size=500),
        batch_size=2),
    event_handler=event_handler,
    num_passes=300)

一开始的时候，我忘记修改cost，用的仍然使用线性回归课程中的cost，即

 cost = paddle.layer.square_error_cost(input=y_predict, label=y)

，运行是正常的。但结果不收敛 检查发现问题后，我把cost修正为现在的样子，即

cost = paddle.layer.classification_cost(input=predict, label=y)

运行出错了。这个错误信息我有点搞不明白，求教：

F1116 05:56:43.303720    11 Matrix.cpp:3626] Check failed: dynamic_cast<CpuIVector*>(&label)
*** Check failure stack trace: ***
    @     0x7f571ade60ed  google::LogMessage::Fail()
    @     0x7f571ade8438  google::LogMessage::SendToLog()
    @     0x7f571ade5bfb  google::LogMessage::Flush()
    @     0x7f571ade930e  google::LogMessageFatal::~LogMessageFatal()
    @     0x7f571ad08152  paddle::CpuMatrix::oneHotCrossEntropy()
    @     0x7f571aae64cf  paddle::CostLayer::forward()
    @     0x7f571ab547a9  paddle::NeuralNetwork::forward()
    @     0x7f571adb62c0  GradientMachine::forwardBackward()
    @     0x7f571a964424  _wrap_GradientMachine_forwardBackward
    @           0x4cb45e  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4ca8d1  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4ca099  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4ca099  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4ca8d1  PyEval_EvalFrameEx
    @           0x4c2765  PyEval_EvalCodeEx
    @           0x4c2509  PyEval_EvalCode
    @           0x4f1def  (unknown)
    @           0x4ec652  PyRun_FileExFlags
    @           0x4eae31  PyRun_SimpleFileExFlags
    @           0x49e14a  Py_Main
    @     0x7f5734ae1830  __libc_start_main
    @           0x49d9d9  _start
    @              (nil)  (unknown)
Aborted

请问是哪里出了问题 还有，我这个例子用来做回归行不行？