GraphTrainableModel#internalPredict works only with one-dimensional predictions

[juliabeliaeva]

It should be possible to make predictions of multiple-dimensions, for example, images, like in the Autoencoder example. However, the linked example fails at https://github.com/JetBrains/KotlinDL/blob/11d4eb175323b61d2e709d4133cbbde49c033746/api/src/main/kotlin/org/jetbrains/kotlinx/dl/api/core/GraphTrainableModel.kt#L698 as predictionsTensor has differrent dimentionality from dst.

[juliabeliaeva]

The simplest fix would be to just linearize predictionsTensor to a FloatArray. This will allow to keep the current api and to make the Autoencoder example work (we already can show FloatArray images in the ImagePanel).
If this is ok, I'll send a pr.

[zaleslaw]

I see this problem, please send a PR to fix that, @juliabeliaeva