RNNの学習において​、多次元入力での学習​は可能でしょうか？

RNN学習において、入力が多次元の場合でも学習可能でしょうか？

http://jp.mathworks.com/help/nnet/ref/layrecnet.html

↑例ではX(特徴量)が1次元であり、試しに

X{1,1}=[0.8147]　のところを　X{1,1}=[0.8147 0.8147]とし、2次元で学習させようとしたところ以下のエラーが出ました。

エラー: nntraining.setup>setupPerWorker (line 64)

Layer states Ai{2,1} and Layer states Ai{1,1}have different numbers of columns.

エラー: nntraining.setup (line 43) [net,data,tr,err] = setupPerWorker(net,trainFcn,X,Xi,Ai,T,EW,enableConfigure);

エラー: network/train (line 335) [net,data,tr,err] = nntraining.setup(net,net.trainFcn,X,Xi,Ai,T,EW,enableConfigure,isComposite);

多次元入力のRNNが学習可能か、また可能であればプログラムの書き方をご教授頂けたら幸いです。 宜しくお願い致します。

[Walter Roberson]

Approximate translation:

In the learning of the RNN, What possible learning of a multi-dimensional input?

In RNN learning, what can be learned, even if input is the case of a multi-dimensional?

http://jp.mathworks.com/help/nnet/ref/layrecnet.html

↑ In the example X (features) is a one-dimensional, to try

The place of X {1,1} = [0.8147] and X {1,1} = [0.8147 0.8147], the following error was trying to learn in a two-dimensional came out.

Error: nntraining.setup> setupPerWorker (line 64)

Layer states Ai {2,1} and Layer states Ai {1,1} have different numbers of columns.

Error: nntraining.setup (line 43) [net, data, tr, err] = setupPerWorker (net, trainFcn, X, Xi, Ai, T, EW, enableConfigure);

Error: network / train (line 335) [net, data, tr, err] = nntraining.setup (net, net.trainFcn, X, Xi, Ai, T, EW, enableConfigure, isComposite);

RNN of multidimensional input or can be learned, also I hope you enjoy teaching how to write, if possible program. Thank you.

[Soya]

thank you for answer.

my question is 'is it possible to train RNN using multi-dimensional input in MATLAB?'.

I tried 2 dimensional-input-training of RNN based in the example. but the error occurred.

[Greg Heath]

The data must be in the form of N pairs of "I"-dimensional "I"nput column vectors and "O"-dimensional "O"utput target column vectors

 [ I N ] = size(input)
 [ O N ] = size(target)

Hope this helps

Greg

Answer 1

[michio]

例えば、X{1,1}=[0.8147; 0.8147] と行数を出力のそれと合わせることで、複数要素の時系列入力を使うことが出来ます。ドキュメンテーションの例ですと、

[X,T] = simpleseries_dataset;
A = cell2mat(X);
AA = [A;A];
XX = mat2cell(AA,2,ones(1,100));

と入力データを作成できますので、試してみてください。

[Xs,Xi,Ai,Ts] = preparets(net,XX,T);
net = train(net,Xs,Ts,Xi,Ai);
view(net)
Y = net(Xs,Xi,Ai);
perf = perform(net,Y,Ts)

また、

>> ntstool

で立ち上がるGUIツールがありますが、ツール上での処理を再現するコードを出力させることもできるのでおすすめです。 Neural Network Time-Series Prediction and Modeling

[Soya]

Michio 様

ありがとうございます。解決いたしました。