Question

我正在开发一个带有张量流和python的深度学习模型：

首先，使用CNN图层，获取功能。
其次，重塑特征映射，我想使用LSTM层。

但是，尺寸不匹配的错误......

ConcatOp：输入的尺寸应匹配：shape[1] = [1200,24]与W_conv1 = weight_variable([1,conv_size,1,12]) b_conv1 = bias_variable([12]) h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1)+ b_conv1) h_pool1 = max_pool_1xn(h_conv1) W_conv2 = weight_variable([1,conv_size,12,24]) b_conv2 = bias_variable([24]) h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2) h_pool2 = max_pool_1xn(h_conv2) W_conv3 = weight_variable([1,conv_size,24,48]) b_conv3 = bias_variable([48]) h_conv3 = tf.nn.relu(conv2d(h_pool2, W_conv3) + b_conv3) h_pool3 = max_pool_1xn(h_conv3) print(h_pool3.get_shape()) h3_rnn_input = tf.reshape(h_pool3, [-1,x_size/8,48]) num_layers = 1 lstm_size = 24 num_steps = 4 lstm_cell = tf.nn.rnn_cell.LSTMCell(lstm_size, initializer = tf.contrib.layers.xavier_initializer(uniform = False)) cell = tf.nn.rnn_cell.MultiRNNCell([lstm_cell]*num_layers) init_state = cell.zero_state(batch_size,tf.float32) cell_outputs = [] state = init_state with tf.variable_scope("RNN") as scope: for time_step in range(num_steps): if time_step > 0: scope.reuse_variables() cell_output, state = cell(h3_rnn_input[:,time_step,:],state) ***** Error In here...

private void dgvFattDdt_SelectionChanged(object sender, EventArgs e)
    {
        if(dgvFattDdt.RowCount == 1)
        {
            dgvFattDdt.ClearSelection();
            return;
        }
        foreach (DataGridViewRow row in dgvFattDdt.SelectedRows)
        {
            //here I save the data of the row into textboxes
            txtColli.Text = row.Cells[1].Value.ToString();
            txtDescMerce.Text = row.Cells[2].Value.ToString();
            txtUM.Text = row.Cells[3].Value.ToString();
            txtIso.Text = row.Cells[4].Value.ToString();
            txtLordo.Text = row.Cells[5].Value.ToString();
            txtTara.Text = row.Cells[6].Value.ToString();
            txtPrezzo.Text = row.Cells[8].Value.ToString();
            cbIVA.Text = row.Cells[10].Value.ToString();
            selectedRow = row.Index;
            ex_Colli = Convert.ToInt32(row.Cells[1].Value.ToString());
            decimal.TryParse(row.Cells[9].Value.ToString(), out ex_Importo);
            int.TryParse(row.Cells[10].Value.ToString(), out ex_Iva);
            //------------------------------
            dgvFattDdt.Rows.RemoveAt(row.Index);
        }
    }

Answer 1

输入到rnn单元格时，输入张量和状态张量的批量大小应相同。

在错误消息中，它显示h3_rnn_input[:,time_step,:]的形状为[71,48]而state的形状为[1200,24]

您需要做的是使第一个维度（batch_size）相同。

如果不打算使用数字71，请检查卷积部分。跨步/填充可能很重要。

Answer 2

您必须考虑正确的号码 h3_rnn_input [：，time_step ,:] 和州以便不按batch_size划分剩余数。（如果有剩余数，则会引发错误）

那么，关于您的代码：

h3_rnn_input[:,time_step,:] has shape of [71,48]
state has shape of [1200,24]

如果我们认为例如batch_size等于90，则：

71/90=0.78 ===> error
1200/90=13.33 ===> error

但是如果我们考虑以下形状和批量大小，那么我们就没有任何问题：

h3_rnn_input[:,time_step,:] has shape of [**60**,48]
state has shape of [**1200**,24]
batch_size=30

然后

60/30=2 ======> ok without any error
1200/30=40 ======> ok without any error

ConcatOp：输入的尺寸应匹配

2 个答案: