0

I didn't find an answer anywhere on internet about why my code is returning an error.

I'm trying to make an LSTM model with tensorflow.net by using the example https://github.com/xuwaters/TensorFlow.NET-Examples/blob/master/src/TensorFlowNET.Examples/ImageProcessing/DigitRecognitionLSTM.cs

But I receive an error (TypeError : If default_name is None then scope is required) when the program is at the line

var (outputs, _, _) = rnn.static_bidirectional_rnn(lstm_fw_cell, lstm_bw_cell, x, dtype: tf.float32);

I tried to put like I saw on internet to define the scope but I still get the error :

using(var scope = tf.name_scope("scope")) { }

I use SciSharp.TensorFlow.Redist 2.11.0 and TensorFlow.NET 0.100.2 nuggets package.

You can find the code of the method here :

        public Graph BuildGraph()
        {
            using(var scope = tf.name_scope("scope"))
            {
                var graph = new Graph().as_default();

                X = tf.placeholder(tf.float32, (-1, timesteps, num_input), name: "X");
                Y = tf.placeholder(tf.float32, (-1, num_classes), name: "Y");

                // Hidden layer weights => 2*n_hidden because of forward + backward cells
                var weights = tf.Variable(tf.random.normal((2 * num_hidden, num_classes)), name: "w");
                var biases = tf.Variable(tf.random.normal(num_classes), name: "b");

                // Unstack to get a list of 'timesteps' tensors of shape (batch_size, num_input)
                var x = tf.split(X, timesteps, 1, "x");

                // Define lstm cells with tensorflow
                // Forward direction cell
                var lstm_fw_cell = new BasicLstmCell(num_hidden, forget_bias: 1.0f);
                // Backward direction cell
                var lstm_bw_cell = new BasicLstmCell(num_hidden, forget_bias: 1.0f);

                // Get lstm cell output
                var (outputs, _, _) = rnn.static_bidirectional_rnn(lstm_fw_cell, lstm_bw_cell, x, dtype: tf.float32);

                // Linear activation, using rnn inner loop last output
                var logits = tf.matmul(outputs.Last(), weights) + biases;
                prediction = tf.nn.softmax(logits);

                // Define loss and optimizer
                loss_op = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
                    logits: logits, labels: Y));
                var optimizer = tf.train.GradientDescentOptimizer(learning_rate: learning_rate);
                train_op = optimizer.minimize(loss_op);

                // Evaluate model (with test logits, for dropout to be disabled)
                var correct_pred = tf.equal(tf.math.argmax(prediction, 1), tf.math.argmax(Y, 1));
                accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32));

                return graph;
            }
        }

Does someone have an idea ? Thank you in advance.

Alex892
  • 1

0 Answers0