theanoを使ってニューラルネットワークの実装でハマりました。

python,機械学習初心者です。
ニューラルネットワークを実装してあるコードをもらい、それを別のものに適応しようと思い変更を加えているのですが、詰まってしまいました。
入力データの設定がおかしいのかなと思ったのですが、わかりませんでした。わかる方がいれば教えていただけると嬉しいです。

発生している問題・エラーメッセージ

Traceback (most recent call last):
  File "homework.py", line 207, in <module>
    loss = func_update(x,t)
  File "/Users/t/.pyenv/versions/anaconda-2.1.0/lib/python2.7/site-packages/theano/compile/function_module.py", line 917, in __call__
    storage_map=getattr(self.fn, 'storage_map', None))
  File "/Users/t/.pyenv/versions/anaconda-2.1.0/lib/python2.7/site-packages/theano/gof/link.py", line 325, in raise_with_op
    reraise(exc_type, exc_value, exc_trace)
  File "/Users/t/.pyenv/versions/anaconda-2.1.0/lib/python2.7/site-packages/theano/compile/function_module.py", line 903, in __call__
    self.fn() if output_subset is None else\
ValueError: number of rows in x (1) does not match length of y (3)
Apply node that caused the error: CrossentropySoftmaxArgmax1HotWithBias(Dot22.0, b, t)
Toposort index: 15
Inputs types: [TensorType(float64, matrix), TensorType(float64, vector), TensorType(int64, vector)]
Inputs shapes: [(1, 3), (3,), (3,)]
Inputs strides: [(24, 8), (8,), (8,)]
Inputs values: [array([[ 0.29324024, -0.40678503,  0.03681703]]), array([0., 0., 0.]), array([1, 0, 0])]
Outputs clients: [[Sum{acc_dtype=float64}(CrossentropySoftmaxArgmax1HotWithBias.0)], [CrossentropySoftmax1HotWithBiasDx(Elemwise{Inv}[(0, 0)].0, CrossentropySoftmaxArgmax1HotWithBias.1, t)], []]

HINT: Re-running with most Theano optimization disabled could give you a back-trace of when this node was created. This can be done with by setting the Theano flag 'optimizer=fast_compile'. If that does not work, Theano optimizations can be disabled with 'optimizer=None'.
HINT: Use the Theano flag 'exception_verbosity=high' for a debugprint and storage map footprint of this apply node.

入力と教師データ

python
1#data_x
2[[5.1 3.5 1.4 0.2]
3 [4.9 3.  1.4 0.2]
4  ......
5 [4.7 3.2 1.3 0.2]
6 [4.6 3.1 1.5 0.2]]
7#data_t
8[[0 0 1]
9 [0 1 0]
10  .....
11 [0 0 1]
12 [0 0 1]]

該当のソースコード

python
1
2from __future__ import print_function
3
4import numpy
5
6import theano
7import theano.tensor as T
8import csv
9
10class Linear(object):
11
12
13    def __init__(self, n_in, n_out):
14
15        # モデルパラメータの初期値を設定
16
17        W_values = numpy.asarray(numpy.random.normal(
18            0, numpy.sqrt(1. / n_in), (n_in, n_out)),
19            dtype=theano.config.floatX)
20        b_values = numpy.zeros((n_out,), dtype=theano.config.floatX)
21
22        self.W = theano.shared(value=W_values, name='W')
23        self.b = theano.shared(value=b_values, name='b')
24
25    def __call__(self, x):
26        return T.dot(x, self.W) + self.b
27
28# MLP モデルを定義しよう
29
30class MnistMLP(object):
31
32    # モデルパラメータ (重み行列やバイアスベクトル) 
33
34    def __init__(self, n_in, n_units, n_out):
35        self.l1=Linear(n_in, n_units)
36        self.l2=Linear(n_units, n_units)
37        self.l3=Linear(n_units, n_out)
38
39    def __call__(self, x):
40        h1 = T.nnet.sigmoid(self.l1(x))
41        h2 = T.nnet.sigmoid(self.l2(h1))
42        return T.nnet.softmax(self.l3(h2))
43# 設定変数
44n_epoch = 10
45n_units = 5
46lr = 0.1
47
48data = numpy.array([[ float(elm) for elm in v] for v in csv.reader(open("fisheriris.csv", "r"))])
49
50data_x,data_t1= numpy.hsplit(data,[4])
51
52data_t = numpy.zeros((data_t1.shape[0],3),dtype = int)
53for i in range(0,data_t.shape[0]):
54    if data_t1[i] == 1:
55        data_t[i,0] = 1
56    elif data_t1[i] == 2:
57        data_t[i,1] = 1
58    else:
59        data_t[i,2] = 1
60
61N_data = data_t.shape[0]
62print(data_t)
63print(data_x)
64
65
66x = T.dmatrix('x')
67t = T.lvector('t')
68
69# モデルを定義しよう
70
71model = MnistMLP(4,n_units,3)
72
73
74y = model(x)
75
76equ_loss = T.mean(T.nnet.categorical_crossentropy(y,t))
77
78
79equ_grad = T.grad(equ_loss,[model.l1.W, model.l1.b,
80                            model.l2.W, model.l2.b,
81                            model.l3.W, model.l3.b])
82
83
84equ_update=[(model.l1.W, model.l1.W - lr*equ_grad[0]),
85            (model.l1.b, model.l1.b - lr*equ_grad[1]),
86            (model.l2.W, model.l2.W - lr*equ_grad[2]),
87            (model.l2.b, model.l2.b - lr*equ_grad[3]),
88            (model.l3.W, model.l3.W - lr*equ_grad[4]),
89            (model.l3.b, model.l3.b - lr*equ_grad[5])]
90
91
92func_update=theano.function(inputs=[x,t],outputs=equ_loss,updates=equ_update)
93
94
95func_loss=theano.function(inputs=[x,t],outputs=equ_loss)
96
97equ_acc=(1.0-T.mean(T.neq(T.argmax(y,axis=1),t)))
98
99func_acc=theano.function(inputs=[x,t],outputs=equ_acc)
100
101
102for epoch in range(1, n_epoch + 1):
103    print('epoch', epoch)
104
105    sum_loss = 0.0
106    sum_accuracy = 0.0
107    perm = numpy.random.permutation(N_data)
108
109    for i in range(0, N_data):
110
111        x = numpy.asarray([data_x[i]])
112        t = numpy.asarray(data_t[i])
113
114
115        loss = func_update(x,t)
116
117
118
119        acc = func_acc(x,t)
120
121
122        sum_loss += float(loss) * len(t)
123        sum_accuracy += float(acc) * len(t)
124
125
126    print('train mean loss={}, accuracy={}'.format(
127        sum_loss / N_data, sum_accuracy / N_data))
128
129
130sum_loss = 0.0
131sum_accuracy = 0.0
132for i in range(0, N_data, batchsize):
133
134    x = numpy.asarray(data_x[perm[i:i + batchsize]])
135    t = data_t[perm[i:i + batchsize]]
136
137    loss = func_loss(x,t)
138
139    acc = func_acc(x,t)
140
141    sum_loss += float(loss) * len(t)
142    sum_accuracy += float(acc) * len(t)
143
144print('test  mean loss={}, accuracy={}'.format(
145    sum_loss / N_data, sum_accuracy / N_data))
146
147