Question

在训练我的神经网络时，它在[float(i) for i in X_Test]给出了这个错误，但是当我将浮动强制转换为X_Train和X_Test时，错误被删除了但我在此处得到了另一个错误ufunc 'add' did not contain a loop with signature matching types dtype('S32') dtype('S32') dtype('S32')代码的一部分x = T.dmatrix('Inputs')

import theano
import theano.tensor as T
import numpy as np
import matplotlib.pyplot as plt
from six.moves import cPickle
import csv, math
from numpy import ceil

# Declare constants
HIDDEN_NEURONS = 50
NUM_CLASSES = 3
NUM_FEATURES = 0
LEARNING_RATE = 0.01
XVAL_SIZE = 0.1
NUM_EPOCHS = 10

raw_data = open("wine.data", 'rb')
temp = list(csv.reader(raw_data, delimiter=',', quoting=csv.QUOTE_NONE))
data = np.array(temp)

np.random.shuffle(data) 

percentage = 75
lines = 0
labels = data[:,0]
x_actual = data[:,0:]

with open("wine.data") as f:
     lines =  len(f.readlines())

X_Train = (x_actual[0:int(np.ceil(lines * (percentage/100)))])
X_Test = (x_actual[int(ceil((lines*(percentage/100)))+1):])

[float(i) for i in X_Train]
[float(i) for i in X_Test]

train_label = labels[:int(ceil(lines*(percentage/100)))]
test_label = labels [int(ceil((lines*(percentage/100)+1))):]

LEARNING_RATE = 0.5
NUM_EPOCHS = 200

rng = np.random.RandomState(1234)

NUM_FEATURES = X_Train.shape[1]

# Declare symbols
x = T.dmatrix('Inputs')
y = T.dmatrix('Outputs')
t = T.dmatrix('Target Values')
Netj = T.dmatrix('Net of hidden layer')
Netk = T.dmatrix('Net of output layer')
Aj = T.dmatrix('Activation of hidden layer')
Wj = np.random.rand(NUM_FEATURES, HIDDEN_NEURONS) * 0.01
Wk = np.random.rand(HIDDEN_NEURONS, NUM_CLASSES) * 0.01
Weights = theano.shared(value=np.concatenate((Wj.ravel(), Wk.ravel()), axis=0),
                    name="Weights ravelled")

# Define equations 
Netj = T.dot(x, Weights[0:NUM_FEATURES * HIDDEN_NEURONS]
         .reshape((NUM_FEATURES, HIDDEN_NEURONS)))


Aj = T.nnet.sigmoid(Netj)

Aj_test = Aj 


y_test = T.nnet.softmax(Aj_test)

cost = T.mean(T.nnet.categorical_crossentropy(y_test, t))


Grads = T.grad(cost, Weights)

# Define Functions

computeCost = theano.function([y_test, t], cost)

forwardProp = theano.function([x], y_test)
forwardProp_test = theano.function([x], y_test)


updates = [(Weights, Weights - LEARNING_RATE * (Grads))]
trainModel = theano.function([x, t], cost, updates=updates)


costs = {'training': list(), 'xval': list()}
for i in range(NUM_EPOCHS):
    print "Epoch number: " + str(i + 1)
    costs['training'].append(trainModel(X_Train, train_label))
    if(i % 10 == 0 and i > 0):
        Test_Result = np.argmax(forwardProp_test(X_Test), axis=1)
        Score = float(len(np.where(Test_Result == train_label)[0])) / float( 
        (train_label.shape[0])) * 100
        print "The model classified with an accuracy of: %.2f" % (
         float(Score)) + "%"

    plt.plot(range(NUM_EPOCHS), costs['training'],
      range(NUM_EPOCHS), costs['xval'])
    plt.show()
    Test_Result = np.argmax(forwardProp(X_Test), axis=1)
    Score = float(len(np.where(Test_Result == train_label)[0])) / float(
   (train_label.shape[0])) * 100
    print "The model performed with an accuracy of: %.2f" % (float(Score)) + "%"

Answer 1

确保您了解

X_Test = x_actual[int(ceil((lines*(percentage/100)))+1):]

产生（（）的外层是不必要的）。它可能是一个数组 - 从错误判断，二维。因此迭代它会产生它的行，1d数组。

如果太大而无法打印整体，请打印其shape和dtype。 type(X_Test)也可能有助于验证它是ndarray。

即使它有效，该表达式也不会将X_Test更改为浮点数。它只是生成一个新的列表或其他内容。

 [float(i) for i in X_Test]

X_Test = X_Test.astype(float)需要生成一个新的浮点数组。

我认为你的第二个错误与第一个错误无关。你永远不会去。这看起来像Theano错误，超出了我的专业领域。

TypeError：只能将length-1数组转换为Python标量神经网络

1 个答案: