chainer sin関数の学習時のエラー

初めに

ディープラーニング初心者です。
chainerでsin関数を学習させています。

内容

input,output数(プログラム内in_units,out_units)を、自由に変更できるプログラムを作成しています。
in_units = 1
out_units = 1
で実行したら、エラーをはかず完了するのですが、
in_units = 2,out_units = 1にしたところ下記のエラーが表示されました。

発生している問題・エラーメッセージ

Exception in main training loop:
Invalid operation is performed in: LinearFunction (Forward)

Expect: x.shape[1] == W.shape[1]
Actual: 1 != 2
Traceback (most recent call last):
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\training\trainer.py", line 343, in run
    update()
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\training\updaters\standard_updater.py", line 240, in update
    self.update_core()
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\training\updaters\standard_updater.py", line 253, in update_core
    optimizer.update(loss_func, *in_arrays)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\optimizer.py", line 874, in update
    loss = lossfun(*args, **kwds)
  File "sinTest.py", line 39, in __call__
    loss= F.mean_squared_error(self.predict(x), y)
  File "sinTest.py", line 46, in predict
    h1 = F.tanh(self.l1(x))
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\link.py", line 287, in __call__
    out = forward(*args, **kwargs)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\links\connection\linear.py", line 183, in forward
    return linear.linear(x, self.W, self.b, n_batch_axes=n_batch_axes)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\functions\connection\linear.py", line 308, in linear
    y, = LinearFunction().apply(args)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\function_node.py", line 307, in apply
    self._check_data_type_forward(in_data)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\function_node.py", line 455, in _check_data_type_forward
    self.check_type_forward(in_type)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\functions\connection\linear.py", line 24, in check_type_forward
    type_check.expect(
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\utils\type_check.py", line 564, in expect
    expr.expect()
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\utils\type_check.py", line 495, in expect
    raise InvalidType(
Will finalize trainer extensions and updater before reraising the exception.
Traceback (most recent call last):
  File "sinTest.py", line 131, in <module>
    trainer.run()
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\training\trainer.py", line 376, in run
    six.reraise(*exc_info)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\six.py", line 719, in reraise
    raise value
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\training\trainer.py", line 343, in run
    update()
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\training\updaters\standard_updater.py", line 240, in update
    self.update_core()
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\training\updaters\standard_updater.py", line 253, in update_core
    optimizer.update(loss_func, *in_arrays)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\optimizer.py", line 874, in update
    loss = lossfun(*args, **kwds)
  File "sinTest.py", line 39, in __call__
    loss= F.mean_squared_error(self.predict(x), y)
  File "sinTest.py", line 46, in predict
    h1 = F.tanh(self.l1(x))
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\link.py", line 287, in __call__
    out = forward(*args, **kwargs)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\links\connection\linear.py", line 183, in forward
    return linear.linear(x, self.W, self.b, n_batch_axes=n_batch_axes)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\functions\connection\linear.py", line 308, in linear
    y, = LinearFunction().apply(args)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\function_node.py", line 307, in apply
    self._check_data_type_forward(in_data)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\function_node.py", line 455, in _check_data_type_forward
    self.check_type_forward(in_type)
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\functions\connection\linear.py", line 24, in check_type_forward
    type_check.expect(
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\utils\type_check.py", line 564, in expect
    expr.expect()
  File "C:\Users\syosh\anaconda3\envs\yoshihiro\lib\site-packages\chainer\utils\type_check.py", line 495, in expect
    raise InvalidType(
chainer.utils.type_check.InvalidType:
Invalid operation is performed in: LinearFunction (Forward)

Expect: x.shape[1] == W.shape[1]
Actual: 1 != 2

該当のソースコード

python
1import numpy as np #配列
2import chainer
3from chainer import cuda, Function, gradient_check, Variable, optimizers, serializers, utils
4from chainer import Link, Chain, ChainList
5import chainer.functions as F
6import chainer.links as L
7import time #時間
8import matplotlib
9matplotlib.use('Agg')
10from matplotlib import pyplot as plt #グラフ
11import os # ディレクトリを操作
12
13# データ
14def get_dataset(N):
15    x = np.linspace(0, 2 * np.pi, N)
16    y = np.sin(x)
17    dataset=[]
18    for xx,yy in zip(x,y):
19        dataset.append((xx,yy))
20    return dataset
21
22# ニューラルネットワーク
23class MyChain(Chain):
24    def __init__(self, in_units=1, n_units=10, out_units=1):
25        super(MyChain, self).__init__(
26             l1=L.Linear(in_units, n_units),
27             l2=L.Linear(n_units, n_units),
28             l3=L.Linear(n_units, n_units),
29             l4=L.Linear(n_units, n_units),
30             l5=L.Linear(n_units, out_units))
31        self.in_units = in_units
32        self.out_units = out_units
33
34
35    def __call__(self, x_data, y_data):
36        x = Variable(x_data.astype(np.float32).reshape(len(x_data),1)) # Variableオブジェクトに変換
37        y = Variable(y_data.astype(np.float32).reshape(len(y_data),1)) # Variableオブジェクトに変換
38        loss= F.mean_squared_error(self.predict(x), y)
39        chainer.reporter.report({
40            'loss': loss
41        }, self)
42        return loss
43
44    def  predict(self, x):
45        h1 = F.tanh(self.l1(x))
46        h2 = F.tanh(self.l2(h1))
47        h3 = F.tanh(self.l3(h2))
48        h4 = F.tanh(self.l4(h3))
49        h5 = self.l5(h4)
50        return h5
51
52    def get_predata(self, x):
53        return self.predict(Variable(x.astype(np.float32).reshape(len(x),self.in_units))).data
54
55from chainer.training import trigger as trigger_module
56class MyChartReport(chainer.training.extension.Extension):
57
58    #trigger = (1, 'epoch')
59
60    def __init__(self,model,N_test=900,trigger = (1, 'epoch')):
61        self.N_test=N_test
62        self.model=model
63        self._trigger = trigger_module.get_trigger(trigger)
64
65    def __call__(self,trainer):
66        if self._trigger(trainer):
67            with chainer.configuration.using_config('train', False):
68                theta = np.linspace(0, 2 * np.pi, self.N_test)
69                sin = np.sin(theta)
70                test = self.model.get_predata(theta)
71                plt.plot(theta, sin, label="sin")
72                plt.plot(theta, test, label="test")
73                plt.legend()
74                plt.grid(True)
75                plt.xlim(0, 2 * np.pi)
76                plt.ylim(-1.2, 1.2)
77                plt.title("sin")
78                plt.xlabel("theta")
79                plt.ylabel("amp")
80                plt.savefig("fig_sin_epoch{}.png".format(trainer.updater.epoch))
81                plt.clf()
82
83# main
84if __name__ == "__main__":
85
86    # 学習データ
87    N = 1000
88    train = get_dataset(N)
89
90    # テストデータ
91    N_test = 900
92    test = get_dataset(N_test)
93
94    # 学習パラメータ
95    batchsize = 10  #バッチサイズ
96    n_epoch = 100   #epoch
97    in_units = 2    #input
98    n_units = 100   #node数
99    out_units = 1   #output
100    gpu=-1
101
102    # モデル作成
103    model = MyChain(in_units, n_units, out_units)
104    optimizer = optimizers.Adam()
105    optimizer.setup(model)
106
107    train_iter = chainer.iterators.SerialIterator(train, batchsize)
108    test_iter = chainer.iterators.SerialIterator(test, batchsize, repeat=False, shuffle=False)
109
110    updater = chainer.training.updaters.StandardUpdater(train_iter, optimizer, device=gpu)
111    trainer = chainer.training.Trainer(updater, (n_epoch, 'epoch'), out="snapshot")
112
113    trainer.extend(chainer.training.extensions.Evaluator(test_iter, model, device=gpu))
114    trainer.extend(chainer.training.extensions.dump_graph('main/loss'))
115
116    # Take a snapshot for each specified epoch
117    frequency = 10
118    trainer.extend(chainer.training.extensions.snapshot(filename="snapshot_cureent"),trigger=(frequency, 'epoch'))
119
120    # Write a log of evaluation statistics for each epoch
121    #trainer.extend(chainer.training.extensions.LogReport())
122
123    trainer.extend(MyChartReport(model,trigger=(10, "epoch")))
124
125    trainer.extend(chainer.training.extensions.LogReport(trigger=(10, "epoch")))
126    trainer.extend(chainer.training.extensions.PrintReport(
127        ["epoch", "main/loss", "validation/main/loss", "elapsed_time"]))
128    #trainer.extend(chainer.training.extensions.ProgressBar())
129    trainer.extend(chainer.training.extensions.PlotReport(['main/loss', 'validation/main/loss'],trigger=(10, "epoch"),filename="loss.png"))
130    trainer.run()
131
132    model.to_cpu()
133    chainer.serializers.save_npz("model.h5",model)