[Feature] ML model caching for reducing prediction overhead (#156)

The caching is now finished. The cache is created in `settings.py` giving us the most flexibility for using it in the future.

The cache is currently updated/accessed by `tasks.py/get_ml_model` which can be called from whatever task needs to access ml models in this way (currently, `predict` and `predict_simple`).

This implementation currently caches all ml models including the relative reasoning. If we don't want this and only want to cache enviFormer, i can change it to that. However, I don't think there is a harm in having the other models be cached as well.

Co-authored-by: Liam Brydon <62733830+MyCreativityOutlet@users.noreply.github.com>
Reviewed-on: enviPath/enviPy#156
Co-authored-by: liambrydon <lbry121@aucklanduni.ac.nz>
Co-committed-by: liambrydon <lbry121@aucklanduni.ac.nz>

epdb/tasks.py

@@ -1,12 +1,19 @@
 import logging
 from typing import Optional
-
+from celery.utils.functional import LRUCache
 from celery import shared_task
 from epdb.models import Pathway, Node, EPModel, Setting
 from epdb.logic import SPathway
 
 
 logger = logging.getLogger(__name__)
+ML_CACHE = LRUCache(3)  # Cache the three most recent ML models to reduce load times.
+
+
+def get_ml_model(model_pk: int):
+    if model_pk not in ML_CACHE:
+        ML_CACHE[model_pk] = EPModel.objects.get(id=model_pk)
+    return ML_CACHE[model_pk]
 
 
 @shared_task(queue="background")
@@ -16,7 +23,7 @@ def mul(a, b):
 
 @shared_task(queue="predict")
 def predict_simple(model_pk: int, smiles: str):
-    mod = EPModel.objects.get(id=model_pk)
+    mod = get_ml_model(model_pk)
     res = mod.predict(smiles)
     return res
 
@@ -51,6 +58,9 @@ def predict(
 ) -> Pathway:
     pw = Pathway.objects.get(id=pw_pk)
     setting = Setting.objects.get(id=pred_setting_pk)
+    # If the setting has a model add/restore it from the cache
+    if setting.model is not None:
+        setting.model = get_ml_model(setting.model.pk)
 
     pw.kv.update(**{"status": "running"})
     pw.save()