Build graph based neural network for time series data using pytorch or DGL

Problem description
• There are total 1000 cases with each case having its own set of node features
• Each case has time series data for 123- time steps
• All the 1000 cases have same edge connectivity and same weight
• Global features of the graph vary for each set of 100 cases.
o For example, global features are same for cases 1-100 and another set of global variable are applied for cases 101-200 and so on…
• Graph based AI model should have following features
o Input: edge connectivity, edge width, node features, global features and 20% of initial time step information input
o Output: predict the 80% of the time step series as output with 90% accuracy
o Graph should be divided into bathes to improve the speed
o Evaluation: For a given edge connectivity, edge width, node features, global features and 20% of initial time step information predict the rest of time step series with 90% accuracy
o It needs to have a feature to visualize the ground truth data in sub plot and predicted data for a given time step in another subplot. This can be used for visual verification.

Please bid only if you have completed any projects in graph based neural network. More details will be provided for successful bidders.