More realistic data

[JackKelly]

• use more appliances from house 1
• appliances from other houses
• train on one set of houses, test on another

[JackKelly]

• increase seq length to more like 3000 (to fit washing machines in there)
• top 5 appliances: kettle, dish washer, htpc (computer), washer dryer, fridge freezer
• train on a mix of appliances from houses 1-4
• validate on appliances from house 5
• need to allow for synonyms of appliance names
• create a single validation dataset which we use across all experiments
• output validation metrics to a CSV file to make it easier to draw new graphs later
• write script to allow multiple experiments to be run automatically and metrics compared (e.g. one plot with all the validation errors on, one plot with all the NILM metrics over time).