forked from enviPath/enviPy
Liam Brydon
901de4640c
**This pull request will need a separate migration pull-request**
I have added an alert box in two places when the user tries to predict with stereo chemistry.
When a user predicts a pathway with stereo chemistry an alert box is shown in that node's hover.
To do this I added two new fields. Pathway now has a "predicted" BooleanField indicating whether it was predicted or not. It is set to True if the pathway mode for prediction is "predict" or "incremental" and False if it is "build". I think it is a flag that could be useful in the future, perhaps for analysing how many predicted pathways are in enviPath?
Node now has a `stereo_removed` BooleanField which is set to True if the Node's parent Pathways has "predicted" as true and the node SMILES has stereochemistry.
<img width="500" alt="{927AC9FF-DBC9-4A19-9E6E-0EDD3B08C7AC}.png" src="attachments/69ea29bc-c2d2-4cd2-8e98-aae5c5737f69">
When a user does a prediction on a model's page it shows at the top of the list. This did not require any new fields as the entered SMILES does not get saved anywhere.
<img width="500" alt="{BED66F12-5F07-419E-AAA6-FE1FE5B4F266}.png" src="attachments/5fcc3a9b-4d1a-4e48-acac-76b7571f6507">
I think the alert box is an alright solution but if you have a great idea for something that looks/fits better please change it or let me know.
Co-authored-by: Tim Lorsbach <tim@lorsba.ch>
Reviewed-on: enviPath/enviPy#250
Co-authored-by: Liam Brydon <lbry121@aucklanduni.ac.nz>
Co-committed-by: Liam Brydon <lbry121@aucklanduni.ac.nz>
91 lines
3.5 KiB
Python
91 lines
3.5 KiB
Python
from collections import defaultdict
|
|
from datetime import datetime
|
|
from tempfile import TemporaryDirectory
|
|
|
|
from django.conf import settings as s
|
|
from django.test import TestCase, override_settings, tag
|
|
|
|
|
|
from epdb.logic import PackageManager
|
|
from epdb.models import EnviFormer, Setting, User
|
|
from epdb.tasks import predict, predict_simple
|
|
|
|
Package = s.GET_PACKAGE_MODEL()
|
|
|
|
|
|
def measure_predict(mod, pathway_pk=None):
|
|
# Measure and return the prediction time
|
|
start = datetime.now()
|
|
if pathway_pk:
|
|
s = Setting()
|
|
s.model = mod
|
|
s.model_threshold = 0.2
|
|
s.max_depth = 4
|
|
s.max_nodes = 20
|
|
s.save()
|
|
pred_result = predict.delay(pathway_pk, s.pk, limit=s.max_depth)
|
|
else:
|
|
pred_result = predict_simple.delay(mod.pk, "C1=CC=C(CSCC2=CC=CC=C2)C=C1")
|
|
_ = pred_result.get()
|
|
return round((datetime.now() - start).total_seconds(), 2)
|
|
|
|
|
|
@tag("slow")
|
|
@override_settings(MODEL_DIR=s.FIXTURE_DIRS[0] / "models", CELERY_TASK_ALWAYS_EAGER=True)
|
|
class EnviFormerTest(TestCase):
|
|
fixtures = ["test_fixtures_incl_model.jsonl.gz"]
|
|
|
|
@classmethod
|
|
def setUpClass(cls):
|
|
super(EnviFormerTest, cls).setUpClass()
|
|
cls.user = User.objects.get(username="anonymous")
|
|
cls.package = PackageManager.create_package(cls.user, "Anon Test Package", "No Desc")
|
|
cls.BBD_SUBSET = Package.objects.get(name="Fixtures")
|
|
|
|
def test_model_flow(self):
|
|
"""Test the full flow of EnviFormer, dataset build -> model finetune -> model evaluate -> model inference"""
|
|
with TemporaryDirectory() as tmpdir:
|
|
with self.settings(MODEL_DIR=tmpdir):
|
|
threshold = float(0.5)
|
|
data_package_objs = [self.BBD_SUBSET]
|
|
eval_packages_objs = [self.BBD_SUBSET]
|
|
mod = EnviFormer.create(self.package, data_package_objs, threshold=threshold)
|
|
|
|
mod.build_dataset()
|
|
mod.build_model()
|
|
mod.evaluate_model(True, eval_packages_objs, n_splits=2)
|
|
|
|
mod.predict("CCN(CC)C(=O)C1=CC(=CC=C1)C")
|
|
|
|
def test_predict_runtime(self):
|
|
with TemporaryDirectory() as tmpdir:
|
|
with self.settings(MODEL_DIR=tmpdir):
|
|
threshold = float(0.5)
|
|
data_package_objs = [self.BBD_SUBSET]
|
|
mods = []
|
|
for _ in range(4):
|
|
mod = EnviFormer.create(self.package, data_package_objs, threshold=threshold)
|
|
mod.build_dataset()
|
|
mod.build_model()
|
|
mods.append(mod)
|
|
|
|
# Test prediction time drops after first prediction
|
|
times = [measure_predict(mods[0]) for _ in range(5)]
|
|
print(f"First prediction took {times[0]} seconds, subsequent ones took {times[1:]}")
|
|
|
|
# Test pathway prediction
|
|
times = [measure_predict(mods[1], self.BBD_SUBSET.pathways[0].pk) for _ in range(5)]
|
|
print(
|
|
f"First pathway prediction took {times[0]} seconds, subsequent ones took {times[1:]}"
|
|
)
|
|
|
|
# Test eviction by performing three prediction with every model, twice.
|
|
times = defaultdict(list)
|
|
for _ in range(
|
|
2
|
|
): # Eviction should cause the second iteration here to have to reload the models
|
|
for mod in mods:
|
|
for _ in range(3):
|
|
times[mod.pk].append(measure_predict(mod))
|
|
print(times)
|