An algorithm that attempts to reload it's model file (if it's been updated) every 5 minutes

model_reloading.py

import Algorithmia
from time import time
import pickle
from src.data import data

client = Algorithmia.client()
DATA_MODEL_DIR = "data://.my/example"
MODEL_NAME = "example.pkl"
TIME_0 = 0
LAST_MODIFIED = ""
MODEL = None
DATA_MODEL_DIR = client.dir(DATA_MODEL_DIR)

def maybe_load():
    global TIME_0, MODEL, DATA_MODEL_DIR, LAST_MODIFIED
    TIME_1 = time()
    # Lets only trigger our maybe load operation every 5 minutes; except for the first time this function is called.
    if TIME_1 - TIME_0 > 300:
        TIME_0 = TIME_1
        current_modified = None
        # Since the only way to get the status of a file in the data API without downloading is via iteration of the
        # directory; this lets us know if the model file has been changed without downloading it.
        for f in DATA_MODEL_DIR.files():
            if f.getName() == MODEL_NAME:
                current_modified = f.last_modified
        # If we haven't loaded a model yet, we should go and do it now
        if MODEL is None and DATA_MODEL_DIR.file(MODEL_NAME).exists():
            local_file = DATA_MODEL_DIR.file(MODEL_NAME).getFile().name
            with open(local_file, 'rb') as f:
                MODEL = pickle.load(f)
        # However if we already have a model; lets make sure that the file's been changed before we redownload and
        # load into memory
        elif current_modified is not None and current_modified != LAST_MODIFIED:
            LAST_MODIFIED = current_modified
            local_file = DATA_MODEL_DIR.file(MODEL_NAME).getFile().name
            with open(local_file, 'rb') as f:
                MODEL = pickle.load(f)


# Dummy apply function expecting a scikit-learn model
def apply(input):
    maybe_load()
    result = MODEL.predict(input)
    return result


maybe_load()

if __name__ == "__main__":
    print(apply([1, 1, 1]))