Constraints:
>to be done in python
>to have run time <30 seconds
Set up:
-Lets say I have an extensive climate dataset consisting of SEASONAL data (rain, temperature, air quality, etc.)
-Lets say I can get this dataset for lots of grid points
-Lets say this data has frequency of hourly
I want to build an algorithm that makes a prediction of this data at some point in the future, using historic data only from that location.
Essentially, I imagine an algorithm working like follows
>You tell the algorithm a latitude/longitude
>The algorithm gets historic data for that latitude/longitude for several features (rain, temp, etc.)
>You tell the algorithm some datetime in the future (reasonably within the next 3 years)
>You tell the algorithm which feature u want to predict (rain, temp, etc.)
>The algorithm, using the historic data, makes a prediction of your feature in the future (predict temperature, predict % chance precip, etc.)
Im wondering how one would approach this in an ML fashion?
The thing is, I know this should be relatively easy, because for example, in the case of temperature, if u literally just took 7 day SMAs from all historic data points for that lat/long, averaged them out and factored in significant temperature trends, you could easily make a great prediction that can be generated super quickly and has great predictive power.... What im wondering if theres a more general ML approach that can be applied when dealing with weather datasets in this fashion?
I try to poke around, but it looks like its all people who are trying to build maximally complex ML algorithms that are unreasonable to be running in the setup i have now (i.e. for an arbitrary lat/long with a short runtime).
Let me know any thoughts u guys have
>to be done in python
>to have run time <30 seconds
Set up:
-Lets say I have an extensive climate dataset consisting of SEASONAL data (rain, temperature, air quality, etc.)
-Lets say I can get this dataset for lots of grid points
-Lets say this data has frequency of hourly
I want to build an algorithm that makes a prediction of this data at some point in the future, using historic data only from that location.
Essentially, I imagine an algorithm working like follows
>You tell the algorithm a latitude/longitude
>The algorithm gets historic data for that latitude/longitude for several features (rain, temp, etc.)
>You tell the algorithm some datetime in the future (reasonably within the next 3 years)
>You tell the algorithm which feature u want to predict (rain, temp, etc.)
>The algorithm, using the historic data, makes a prediction of your feature in the future (predict temperature, predict % chance precip, etc.)
Im wondering how one would approach this in an ML fashion?
The thing is, I know this should be relatively easy, because for example, in the case of temperature, if u literally just took 7 day SMAs from all historic data points for that lat/long, averaged them out and factored in significant temperature trends, you could easily make a great prediction that can be generated super quickly and has great predictive power.... What im wondering if theres a more general ML approach that can be applied when dealing with weather datasets in this fashion?
I try to poke around, but it looks like its all people who are trying to build maximally complex ML algorithms that are unreasonable to be running in the setup i have now (i.e. for an arbitrary lat/long with a short runtime).
Let me know any thoughts u guys have
