Many changes have been done to the NILMTK-API. The code needs to be changed slightly in order to make it run with the API. Refer to the NILMTK-contrib about the latest documentation.
In this repository you can find the notebooks that are associated with the paper results of the NILMTK's Buildsys 2019 paper. The notebooks demonstrate the power of the new API.
The algorithms used in the paper are as follows
- Mean Algorithm
- Hart's Algorithm
- Combinatorial Optimization
- Exact FHMM
- Discriminative Sparse Coding
- Additive FHMM
- Additive FHMM with SAC (Signal Aggregate Constraints)
- Denoising Auto Encoder
- RNN
- WindowGRU
- Seq2Point
- Seq2Seq
Algorithms such as AFHMM, AFHMM with SAC and Discriminative Sparse Coding are CPU intensive. All the neural networks are GPU intensive, so the a single experiment had to be run of different types of machines. All the CPU intensive algorithms were run on very powerful CPU system and every other algorithm was run on a system with a GPU. So, for every experiment we have two different notebooks - one for CPU algorithms and another for everything else.