search

dafrie / lstm-load-forecasting

Electricity load forecasting with LSTM (Recurrent Neural Network)
MIT License
164 stars 77 forks source link

weather_i #2

Open samebohon opened 5 years ago

samebohon commented 5 years ago

Thanks so much for sharing.

I have a really basic question. What are the "weather_i" variables are they dummy variables or prehaps some other weather observation type supplementing the temperature data?

Thanks

samebohon

dafrie commented 5 years ago

Sorry, my documentation on the public repository is not great and I had to check my term paper I wrote from back then:

"For each city, the temperature (◦F) and a qualitative weather categorization (bad, neutral, nice) has been collected. In total, there are 32 features describing weather."

So its actually One-Hot-Encoding of a 3 category variable and the original data was the "icon" from the Dark Sky API, with the mapping defined in the config.json:

{
   ...
  "iconmap": {
    "clear-day": 1,
    "clear-night": 2,
    "rain": 3,
    "snow": 3,
    "sleet": 3,
    "wind": 3,
    "fog": 3,
    "cloudy": 3,
    "partly-cloudy-day": 1,
    "partly-cloudy-night": 2
  }
}

Looking back at this, it seems actually quite random so I wouldn't interpret much into this variable ;)

samebohon commented 5 years ago

Hi there,

Thanks for this. After I sent the query I actually figured it out but it is great to have your confirmation.

I am intrigued, are you working in energy modelling at present, or did you leave energy behind after this?

Thanks