I/O Control

It is possible to save and load both the fitted model and the generated sequence(s) of precipitation and temperature, although this functionality has not been explicitly included in swxg to give greater control to the user over how these objects integrate with their own workflows. The following provides some examples of how the the model and the generated data could be saved and loaded.

Let’s assume we have two objects we’re trying to save:

  • model, which is the fitted model

  • wx, which is the dataframe that contains all of the generated sequences of precipitation and temperature across all sites by model

Saving

Saving model is simple with the pickle module (which is a standard Python library):

# import pickle
import pickle

# save `model` using pickle
filename = "/path/to/dir/swxg_model.pkl"
with open(filename, "wb") as f:
    pickle.dump(model, f)

Saving wx is also simple since it is a Pandas dataframe, which means all methods that work to save dataframes work here as well:

filename = "/path/to/dir/wx"

# save `wx` as .csv (this will not save information about the columns like typing)
extension = ".csv"
wx.to_csv(filename + extension, index=False)

# save `wx` as .pkl (import pickle to do this)
import pickle
extension = ".pkl"
wx.to_pickle(filename + extension)

# save `wx` as `.parquet` (import a Parquet library like pyarrow to do this)
import pyarrow
extension = ".parquet"
wx.to_parquet(filename + extension)

Loading

Loading the saved model can also be done with pickle:

with open("/path/to/dir/swxg_model.pkl", "rb") as f:
    loaded_model = pickle.load(f)

Both model and loaded_model will be identical instances of SWXGModel, and thus loaded_model can be used to generate new sequences of precipitation and temperature based on the inputs given to model.

Loading the generated dataframe should also be done through the native Pandas methods:

# need to recover column type information
wx_csv = pd.read_csv("/path/to/dir/wx.csv")
wx_csv = wx_csv.astype({"SITE": str, "YEAR": int, "MONTH": int, "PRECIP": float, "TEMP": float})

# no need to recover column type information, already saved
wx_pkl = pd.read_pickle("wx.pkl")

# no need to recover column type information, already saved
wx_parquet = pd.read_parquet("wx.parquet")

By reading this data back in with the native Pandas methods, all three saved generated sequences should be identical to the original wx dataframe.