The exported test vs prediction plot gives unexpected/erroneous results. Like the following exported image shows.

This does not happen however when using python directly without sockets. Following image is running on similar data.

Perhaps using Plotly could solve above problem