input must be a scalar, but it has shape [1]

[zhengdi1992]

in tensorflow javacpp..
the placeholder was define by

self.drop = tf.placeholder(tf.float32,name='drop')
self.phase = tf.placeholder(tf.bool,name='x_phase')

in java, how to build the tensor?
i have try like this

tensorflow.Tensor drop = new tensorflow.Tensor(tensorflow.DT_FLOAT, new TensorShape(new long[]{1}));
FloatRawIndexer dropIndexer = drop.createIndexer();
dropIndexer.put(0, 0.5f);

but occur the error when session run..

input must be a scalar, but it has shape [1]
	 [[Node: fc_1layer/cond/sub/Switch = Switch[T=DT_FLOAT, _class=["loc:@drop"], _device="/job:localhost/replica:0/task:0/device:CPU:0"](_arg_drop_0_1, _arg_x_phase_0_0)]]

i think was the new TensorShape(new long[]{1}) should not like that, but a scalar shape?
but how build it? if use new TensorShape(), the data cannot fill into the tensor..dropIndexer.put(0, 0.5f); will occur ArrayIndexOutOfBoundsException exeption

[saudet]

TensorFlow considers arrays that have a number of dimensions equal to 0 as scalars, so something like new TensorShape(new long[]{}) works for that.

[zhengdi1992]

but if that new TensorShape(new long[]{}), how can assign the value to the tensor?
if in python, can use like that and works well.. how to translate to javacpp?

drop = sess.graph.get_tensor_by_name('drop:0')
op = sess.graph.get_tensor_by_name('str_output:0')
sess.run(op, {drop: 0.5})

i got the ArrayIndexOutOfBoundsException exception, by use that..when drop.createIndexer();

tensorflow.Tensor drop = new tensorflow.Tensor(tensorflow.DT_FLOAT,
            new TensorShape(new long[]{}));
FloatRawIndexer dropIndexer = drop.createIndexer();
dropIndexer.put(0, 0.5f);

[saudet]

There was an issue previously with scalars like that, but it should work with the latest release.

[zhengdi1992]

the version is that?

<dependency>
    <groupId>org.bytedeco.javacpp-presets</groupId>
    <artifactId>tensorflow</artifactId>
    <version>1.12.0-1.4.4</version>
</dependency>

[saudet]

No, 1.13.1-1.5, see: http://bytedeco.org/download/
The groupId and package names changed a bit:
http://bytedeco.org/news/2019/04/11/beyond-java-and-cpp/

[zhengdi1992]

upgrade it and works well..
yep, changed a bit...hahaha..

[saudet]

I've added a couple more overloads for Tensor.create() so we can now do something like this instead:

        Tensor scalar = Tensor.create(new float[] {42});
        Tensor vector = Tensor.create(new float[] {0, 1}, 2);