Open martiansideofthemoon opened 7 years ago
Minimum how many factors should we consider?
I think we should start with 4-5 parameters atleast. I can think of time of the year, location as very important factors to begin with.
Time of the year, location - altitude, distance to sea, terrain type(grassland, desert , ... ), recent weather history for that place(wind speed, humidity, temperature, rain), weather history of that place at that time of the year in previous years are relevant factors for deciding weather.
That seems correct @tejus-gupta . Let's aim at each making PRs where we add a link to a good dataset for our usage. Here is good place to start https://www.quora.com/How-can-I-get-free-data-sets-for-weather-forecasting
finally what all parametres to consider, if we are going to use the wunderground api? max. I think max-min temp and humidity, fog and rain are good to start with. we can take wind direction and speed also. regarding the output, what all parametres should we present as output, i.e. how to parametrize and present the output?
Have you seen the sample data in the wunderground api? let's condition one by one, I'd say take the date
, humidity
, max-min temperature
and fog to predict amount of rainfall. Use a simple feedforward network.
Hey @martiansideofthemoon , I want to work on this particular project but I'm facing a problem. I've written a python script to extract just date, humidity, min temp, max temp, fog and rainfall of each day from the json file(As provided in one of the commits by @ronalddas), which will serve as training set for our neural network. However we'll need more of the training data. I'm facing a problem(500 Internal Server Error) registering for wunderground. It would be really helpful if someone can either provide me their api key or provide me the data of at least past 100 days of any city.
@ronalddas could you help him out?
I now want to use TensorFlow to predict the precipitation on the grid, do you have any good suggestions, do you think it is using cnn or rnn, what are you doing now, and hope to explore each other
@zengxiaoqing CNNs sound like a good idea
This includes quantifying each of the chosen parameters. We need to feed numbers into the Neural Networks eventually.