Integrate NCES in ontolearn-web service and fix dllearner binding script in examples #450

Jean-KOUAGOU · 2024-10-15T14:24:12Z

This pull request addresses the issue#408: #408, and updates the dllearner binder script such that positive and negative examples initially given here: https://github.com/dice-group/Ontolearn/blob/develop/examples/dl_learner.py#L39 are first sent to the PosNegLPStandard class as required by the binder and most class expression learners, inc. Drill, EvoLearner, CELOE, CLIP, etc

Please review and merge

…o ontolearn-webservice

Demirrr · 2024-10-16T09:14:28Z

ontolearn/concept_learner.py

@@ -1572,7 +1572,7 @@ def best_hypotheses(self, n=1) -> Union[OWLClassExpression, Iterable[OWLClassExp
        elif len(self.best_predictions) == 1 or n == 1:
            return self.best_predictions[0].concept
        else:
-            return self.best_predictions[:n]
+            return [best.concept for best in self.best_predictions[:n]]


Why do we need this converation ?

By default, other concept learners, including Drill output a list of concepts, not nodes. This is why I am converting. Moreover, NCES uses its own node class for the moment. In the future, if necessary, we might investigate the use of OENode which is currently tailored towards search-based approaches.

We cannot return OENode. Since we learn OWL Class expressions, return results must fulfill Union[OWLClassExpression, Iterable[OWLClassExpression]].

ontolearn/scripts/run.py

examples/dl_learner.py

Demirrr · 2024-10-17T08:04:49Z

Could you please ensure that no fixed paths are being used?
The following lines must not contain only fixed paths

Ontolearn/ontolearn/scripts/run.py

Line 59 in 49427ed

parser.add_argument("--path_of_nces_embeddings", type=str, default="./NCESData/")
Ontolearn/ontolearn/scripts/run.py

Line 92 in 49427ed

if not os.path.exists(f"./NCESData/{kb_name}/embeddings/ConEx_entity_embeddings.csv"):
Ontolearn/ontolearn/scripts/run.py

Line 106 in 49427ed

embeddings_path += f"{kb_name}/embeddings/ConEx_entity_embeddings.csv"

There is no reason to force a user to use conex embeddings
Ontolearn/ontolearn/scripts/run.py

Line 149 in 49427ed

path_of_embeddings=get_embedding_path("https://files.dice-research.org/projects/NCES/NCES_Ontolearn_Data/NCESData.zip", args.path_knowledge_base),

owl_learner must follow the same interface.

Ontolearn/ontolearn/scripts/run.py

Line 195 in 49427ed

if isinstance(owl_learner, NCES):

…ghts do not exist

….fit(pos, neg)

Jean-KOUAGOU · 2024-10-17T10:51:49Z

, 2., 3., 4.

parser.add_argument("--path_of_nces_embeddings", type=str, default="./NCESData/") was removed. It now takes input from the user (same as Drill data["path_embeddings"])

if not os.path.exists(f"./NCESData/{kb_name}/embeddings/ConEx_entity_embeddings.csv"). Same as above. The function has been modified, and takes input from the user. It is only when the path provided by the user does not exist that it looks for embeddings online. This is because NCES needs embeddings to run, not like Drill which can run without embeddings.

owl_learner must follow the same interface. This has been fixed. No check whether it is an NCES instance. NCES takes PosNegLPStandard as the other learners now.

Note:

If I completely remove the possibility to download embeddings for NCES, the user will have to download them manually in case he does not have any emeddings and still needs to test NCES. As it currently is, the user can provide any embeddings and they will be used. But feel free to request that I remove any possibility to download data.

Jean-KOUAGOU · 2024-10-17T10:56:55Z

Another Note

NCES is trained from scratch if the provided path for pretained models does not contain any files

Demirrr · 2024-10-17T12:55:12Z

Can you please follow the enumeration ? and fix the each point individually.

, 2., 3., 4.

What does this enumeration mean ? Can you please nextime explicitly state whether the required is done or not?

This function is still in the script. This needs to be removed.

def get_embedding_path(ftp_link: str, kb_path_arg: str)->str:
        return f"./NCESData/{kb_name}/embeddings/ConEx_entity_embeddings.csv"

Please follow the same structure of preparting an owl reasoner as we have done for DRILL and TDL

Demirrr · 2024-10-17T12:56:22Z

Is still not being fixed

Ontolearn/ontolearn/scripts/run.py

Line 59 in 49427ed

parser.add_argument("--path_of_nces_embeddings", type=str, default="./NCESData/")

Jean-KOUAGOU · 2024-10-17T13:27:20Z

Is still not being fixed

Ontolearn/ontolearn/scripts/run.py

Line 59 in 49427ed

parser.add_argument("--path_of_nces_embeddings", type=str, default="./NCESData/")

You are probably looking at an old commit. New link here: https://github.com/dice-group/Ontolearn/blob/nces_web_and_fix_dllearner_binder/ontolearn/scripts/run.py#L55

Jean-KOUAGOU · 2024-10-17T13:29:31Z

I replied for 1., 2., 3., 4.

They have been resolved but with notes awaiting for your confirmation. I will write notes down here again.

parser.add_argument("--path_of_nces_embeddings", type=str, default="./NCESData/") was removed. It now takes input from the user (same as Drill data["path_embeddings"])

if not os.path.exists(f"./NCESData/{kb_name}/embeddings/ConEx_entity_embeddings.csv"). Same as above. The function has been modified, and takes input from the user. It is only when the path provided by the user does not exist that it looks for embeddings online. This is because NCES needs embeddings to run, not like Drill which can run without embeddings.

owl_learner must follow the same interface. This has been fixed. No check whether it is an NCES instance. NCES takes PosNegLPStandard as the other learners now.

Note 1:

If I completely remove the possibility to download embeddings for NCES, the user will have to download them manually in case he does not have any emeddings and still needs to test NCES. As it currently is, the user can provide any embeddings and they will be used. But feel free to request that I remove any possibility to download data.

Note 2:

NCES is trained from scratch if the provided path for pretained models does not contain any files

Demirrr · 2024-10-17T13:59:21Z

They have been resolved but with notes awaiting for your confirmation

If they have been resolved, please push them.

examples/train_nces.py

ontolearn/scripts/run.py

Demirrr · 2024-10-17T14:03:20Z

ontolearn/scripts/run.py

+                    num_predictions=64
+                   )
+    # (2) Either load the weights of NCES or train it.
+    if data.get("path_to_pretrained_nces", None) and os.path.isdir(data["path_to_pretrained_nces"]) and glob.glob("*.pt"):


Please do not use glob.plot with ** * **. Input items must be fixed

Demirrr · 2024-10-17T15:10:58Z

@Jean-KOUAGOU if it makes the integration easier, we can ask a user to provide the paths of csv file containing embeddings and pretrained NCES weights. Without these two arguments at the start, NCES would return Top concept for any learning problem

…rovided path does not contain any embeddings. ontolearn-webservice will now require path to existing embeddings for NCES to run

examples/concept_learning_cv_evaluation.py

examples/dl_learner.py

Jean-KOUAGOU · 2024-10-23T07:31:48Z

Hi. If a comma or import is missing, it must have been a typo. Please feel free to add them.

…

On Wed, Oct 23, 2024, 07:13 Luke Friedrichs ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In examples/dl_learner.py <#450 (comment)>: > @@ -36,9 +38,14 @@ p = examples['positive_examples'] n = examples['negative_examples'] - best_pred_celoe = celoe.fit(pos=p, neg=n, max_runtime=1).best_hypothesis() + positives = {OWLNamedIndividual(IRI.create(i)) for i in p} + negatives = {OWLNamedIndividual(IRI.create(i)) for i in n} + + lp = PosNegLPStandard(pos=positives, neg=positives) Import for this is missing — Reply to this email directly, view it on GitHub <#450 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AK4X4ETCP32LHK4MAXPQ3YTZ44V67AVCNFSM6AAAAABP7JIJI2VHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDGOBXGE3DKMJXHE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

ontolearn/concept_learner.py

Jean-KOUAGOU added 2 commits October 15, 2024 15:47

fix example script to run dllearner from source, and integrate NCES t…

c4c6a9a

…o ontolearn-webservice

update documentation

5824bbc

Jean-KOUAGOU requested a review from Demirrr October 15, 2024 14:24

github-actions bot temporarily deployed to pull request October 15, 2024 14:30 Inactive

Demirrr requested changes Oct 16, 2024

View reviewed changes

Set NCES to train from scratch when no pretrained model is available

49427ed

github-actions bot temporarily deployed to pull request October 16, 2024 16:27 Inactive

In ontolearn-webservice, NCES will first be trained if pretrained wei…

5aa443e

…ghts do not exist

github-actions bot temporarily deployed to pull request October 17, 2024 09:31 Inactive

update nces_notebooks with new argument names in NCES instantiation

a2a40aa

github-actions bot temporarily deployed to pull request October 17, 2024 09:50 Inactive

NCES can now fit on a PosNegLPStandard object

3cd8ee8

github-actions bot temporarily deployed to pull request October 17, 2024 10:30 Inactive

update NCES notebooks in examples to use nces.fit(lp) instead of nces…

7485355

….fit(pos, neg)

github-actions bot temporarily deployed to pull request October 17, 2024 10:38 Inactive

NCES.fit() now takes a PosNegLPStandard object

d720568

github-actions bot temporarily deployed to pull request October 17, 2024 10:47 Inactive

fix NCES bug in scripts/run.py

cb82167

github-actions bot temporarily deployed to pull request October 17, 2024 11:11 Inactive

Jean-KOUAGOU requested a review from Demirrr October 17, 2024 13:34

Demirrr requested changes Oct 17, 2024

View reviewed changes

remove possibility to automatically download embeddings for NCES if p…

43f94d1

…rovided path does not contain any embeddings. ontolearn-webservice will now require path to existing embeddings for NCES to run

github-actions bot temporarily deployed to pull request October 17, 2024 15:29 Inactive

Merge branch 'develop' into nces_web_and_fix_dllearner_binder

e4b59ec

github-actions bot temporarily deployed to pull request October 18, 2024 14:06 Inactive

LckyLke reviewed Oct 23, 2024

View reviewed changes

examples/concept_learning_cv_evaluation.py Outdated Show resolved Hide resolved

LckyLke reviewed Oct 23, 2024

View reviewed changes

examples/dl_learner.py Show resolved Hide resolved

alkidbaci reviewed Oct 24, 2024

View reviewed changes

ontolearn/concept_learner.py Outdated Show resolved Hide resolved

Fixed bugs

cf238d4

github-actions bot temporarily deployed to pull request October 24, 2024 13:35 Inactive

refactoring due to removal of fixed paths

0b7da0c

github-actions bot temporarily deployed to pull request October 24, 2024 14:29 Inactive

Demirrr merged commit 4f0b2a5 into develop Oct 25, 2024
4 checks passed

Demirrr deleted the nces_web_and_fix_dllearner_binder branch October 25, 2024 08:57

github-actions bot temporarily deployed to pull request October 25, 2024 08:58 Inactive

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Integrate NCES in ontolearn-web service and fix dllearner binding script in examples #450

Integrate NCES in ontolearn-web service and fix dllearner binding script in examples #450

Jean-KOUAGOU commented Oct 15, 2024

Demirrr Oct 16, 2024

Jean-KOUAGOU Oct 16, 2024

Demirrr Oct 16, 2024

Demirrr commented Oct 17, 2024

Jean-KOUAGOU commented Oct 17, 2024 •

edited

Loading

Jean-KOUAGOU commented Oct 17, 2024

Demirrr commented Oct 17, 2024

Demirrr commented Oct 17, 2024

Jean-KOUAGOU commented Oct 17, 2024

Jean-KOUAGOU commented Oct 17, 2024 •

edited

Loading

Demirrr commented Oct 17, 2024

Demirrr Oct 17, 2024

Demirrr commented Oct 17, 2024

Jean-KOUAGOU commented Oct 23, 2024 via email

Integrate NCES in ontolearn-web service and fix dllearner binding script in examples #450

Integrate NCES in ontolearn-web service and fix dllearner binding script in examples #450

Conversation

Jean-KOUAGOU commented Oct 15, 2024

Demirrr Oct 16, 2024

Choose a reason for hiding this comment

Jean-KOUAGOU Oct 16, 2024

Choose a reason for hiding this comment

Demirrr Oct 16, 2024

Choose a reason for hiding this comment

Demirrr commented Oct 17, 2024

Jean-KOUAGOU commented Oct 17, 2024 • edited Loading

Note:

Jean-KOUAGOU commented Oct 17, 2024

Another Note

Demirrr commented Oct 17, 2024

Demirrr commented Oct 17, 2024

Jean-KOUAGOU commented Oct 17, 2024

Jean-KOUAGOU commented Oct 17, 2024 • edited Loading

Note 1:

Note 2:

Demirrr commented Oct 17, 2024

Demirrr Oct 17, 2024

Choose a reason for hiding this comment

Demirrr commented Oct 17, 2024

Jean-KOUAGOU commented Oct 23, 2024 via email

Jean-KOUAGOU commented Oct 17, 2024 •

edited

Loading

Jean-KOUAGOU commented Oct 17, 2024 •

edited

Loading