CODEWITHSUNDEEP
pythonnumpyneural-networktheano
Engineer
Engineer

Reputation: 75

TypeError: only length-1 arrays can be converted to Python scalars Neural network

When training my Neural Network, It gives me this error at [float(i) for i in X_Test], but when I remove the float casting to the X_Train and X_Test the error is removed but I get another error instead ufunc 'add' did not contain a loop with signature matching types dtype('S32') dtype('S32') dtype('S32') at this part of code 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)) + "%"

Upvotes: 0

Views: 412

Answers (1)

hpaulj
hpaulj

Reputation: 231385

Make sure you understand what 

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

produces (the outer layer of () is unnecessary). It probably is an array - and judging from the error, 2d dimensional. So iterating on it produces its rows, 1d arrays.

If it is too large to print whole, print its shape and dtype. type(X_Test) might also help to verify that it is an ndarray.

Even if it worked, this expression would not change X_Test to floats. It just produces a new list of something or other.

 [float(i) for i in X_Test]

X_Test = X_Test.astype(float) is required to produce a new array of floats.

I think your 2nd error has nothing to do with the first one. You just never go to it. That looks like a Theano error, which is outside my area of expertise.

Upvotes: 1

Related Questions