Adding entity resolution feature to SANSA-ML. #24

amrit1210 · 2019-10-27T16:27:39Z

The feature uses minhashLSH based approach to identify and link similar entities in two datasets

…hLSH based approach to identify and link similar entities in two datasets

GezimSejdiu

thanks a lot for your contributions. I just went through it and added some comments. Please, try to resolve them or add your comments so we can discuss them in more detail. In particular, do:

Please, have a look that we do follow the official Scala-style guide and it should be adapted accordingly into your code as well (see my comments on the review). If you want to validate it, just run mvn scalastyle:check and if that pass you are okay to push changes.
Try to write code once. There are many duplicates between both classes and it would be better if we can write reusable, generic code and/or calling existing methods instead.
No unit-test provided. We should have some basic unit-test where individual units/ components of the component are tested.

Looking forward to resolving those issues and move forward on mergeing the PR.

Best regards,

...la/net/sansa_stack/ml/spark/entity_resolution/EntityResolution_RDFData_CountVectorizer.scala

GezimSejdiu · 2019-11-17T20:42:55Z

...la/net/sansa_stack/ml/spark/entity_resolution/EntityResolution_RDFData_CountVectorizer.scala

+
+    // Compare the predicates for matched entities by subject
+    val refined_data_pred = get_similar_predicates(ds_subjects)
+    val ds_predicates = refined_data_pred.repartition(400).persist(StorageLevel.MEMORY_AND_DISK)


Same here. We should remove any hard-coded code.

GezimSejdiu · 2019-11-17T20:45:49Z

...la/net/sansa_stack/ml/spark/entity_resolution/EntityResolution_RDFData_CountVectorizer.scala

+  def get_entity_profiles(triplesRDD: RDD[Triple]): RDD[(String, String)] = {
+
+    //predicates to be filtered out from triples
+    val removePredicates: List[String] = List("owl:sameas", "wikiPageID", "wikiPageRevisionID", "wikiPageRevisionLink",


I see that it is quite static? What will happen if there is another dataset to be used? Can we add them to a specific configuration file? or even an input (parameter, file) which allows users to define their predicates they want to filter them out. We could make it even more extensible by allowing users to add them as RDF triples (we could define as simple RDF graph which contains these predicates to filter out, but this is optional - the RDF graph and not moving them out of the hard-coded variable).

GezimSejdiu · 2019-11-17T20:46:49Z

...la/net/sansa_stack/ml/spark/entity_resolution/EntityResolution_RDFData_CountVectorizer.scala

+
+    //Get subject data and tokenise it
+    val ent_sub1 = entites_RDD1.map(f => { (f._1, f._1.split("_")) })
+    val part_rdd1 = new RangePartitioner(400, ent_sub1)


Again, this number is quite static and we should find a way how to define it (if it is needed).

GezimSejdiu · 2019-11-17T20:48:36Z

...la/net/sansa_stack/ml/spark/entity_resolution/EntityResolution_RDFData_CountVectorizer.scala

+
+    //Calculate TruePostives for precision, recall and f1-measure
+    val truePositives = actual_rdd.intersection(predicted_rdd).count
+    println("***************************************************************************************")


No printlns statements on API. We should move them to SANSA-Examples.

Do the same for the rest as well.

GezimSejdiu · 2019-11-17T20:50:50Z

...in/scala/net/sansa_stack/ml/spark/entity_resolution/EntityResolution_RDFData_HashingTF.scala

+ * Dataset1 has 47 million entities, approx 19GB
+ * Dataset2 has 19 million entities, approx 5.8GB
+ * */
+class EntityResolution_RDFData_HashingTF(spark: SparkSession, triplesRDD1: RDD[Triple], triplesRDD2: RDD[Triple],


Same goes here (regarding the scala-style naming conventions).

GezimSejdiu · 2019-11-17T20:52:22Z

...in/scala/net/sansa_stack/ml/spark/entity_resolution/EntityResolution_RDFData_HashingTF.scala

+   * where subject is the key and the group of paired predicate:object forms the value
+   * ex:- (Budapest_City, areaCode:1 , country:Hungary)
+   */
+  def get_entity_profiles(triplesRDD: RDD[Triple]): RDD[(String, String)] = {


Is there any difference in this method from the class above? If not, please consider writing code-once and reuse it when it's needed. I suggest that you merge them into e.g. Utils class or similar and do not keep t wo identic code snippets into two different classes of the same package.

And I see that there are also other methods which are the same in both classes. Please, merge them and define them only once. I will not comment on similar methods twice. Consider, the same messages above for the same methods.

GezimSejdiu

Hi @amrit1210 ,

many thanks for your updates and contributions. Please, go through my comments and try to resolve them.
Important points to be considered:

make it scala-style compliant
remove files from the main package into resources
provide basic unit-test and move out main method into SANSA-Examples
the other comments

Best regards,

GezimSejdiu · 2020-01-08T21:27:20Z

sansa-ml-spark/src/main/scala/net/sansa_stack/ml/spark/entity_resolution/Commons.scala

-                                         teacher: DataFrame, threshold_subject: Double, jsimilarity_predicate: Double,
-                                         threshold_object: Double, vocab_size: Long, output_path: String) extends Serializable {
+abstract class Commons(val spark: SparkSession, val sourceData1: RDD[Triple], val sourceData2: RDD[Triple],
+                                         val teacher: DataFrame, val thresholdSubject: Double, val jsimilarityPredicate: Double,


Why teacher? Are we making this generic enough which will accept any other type (RDF) dataset? I think we should remove such a misleading name. Of course, in your use-case, it makes totally sense, but here we should keep it more generic, as we aim to provide it as an API within SANSA.

teacher is basically the labelled data to which we compare our predictions later on to evaluate recall and precision values. Yes, our algorithm can accept any other RDF dataset but n order to evaluate how well the entities are resolved, we would need the gold standards i.e. teacher in our case. I can change it to labels if you suggest.

Right. I think you should include that DF as part of the algorithm. If you use validation data just for the sake of prediction and not training, I think we should generalize it to accept any other validation data. So, better remove it from the main algorithm and just add it when running the example.

GezimSejdiu · 2020-01-08T21:31:52Z

sansa-ml-spark/src/main/scala/net/sansa_stack/ml/spark/entity_resolution/Commons.scala

-class EntityResolution_RDFData_CountVectorizer(spark: SparkSession, triplesRDD1: RDD[Triple], triplesRDD2: RDD[Triple],
-                                         teacher: DataFrame, threshold_subject: Double, jsimilarity_predicate: Double,
-                                         threshold_object: Double, vocab_size: Long, output_path: String) extends Serializable {
+abstract class Commons(val spark: SparkSession, val sourceData1: RDD[Triple], val sourceData2: RDD[Triple],


Why Commons? and why abstract? To my understanding, The abstract methods of an abstract class are those methods that do not contain any implementation. and in your case, your methods are fully implemented. I will suggest that you create a package utils which then contains such methods (grouped by their functionality into separate Scala objects).

The Commons class contains the abstract method vectorise which has seperate implemntations in the two subclass that inherit this class. That's why it is structured in this way. If you still think I should do it in some other better way. PLease suggest.

Not sure, but feel free to keep it as you have if you think that is the right way to do. Later, we can refactor it if needed.

sansa-ml-spark/src/main/scala/net/sansa_stack/ml/spark/entity_resolution/Commons.scala

GezimSejdiu · 2020-01-08T21:37:50Z

sansa-ml-spark/src/main/scala/net/sansa_stack/ml/spark/entity_resolution/Commons.scala

@@ -35,22 +39,22 @@ class EntityResolution_RDFData_CountVectorizer(spark: SparkSession, triplesRDD1:
  def run(): RDD[(String, String)] = {


As I already mentioned it on my previous review, we should get rid of any main method into API. We could add unit-tests (by the way, I do not see any unit tests here, we usually should have at least some coverage with the functionality before we merge the PR) which cover some of the functionality and the main run we should consider moving it to the SANSA-Examples into ml examples.

This method is required, so as to accomplish the 3-step procedure for our algorithm. As the output from each step becomes the input for the next, being a sequential process. We are including unit tests separately.

GezimSejdiu · 2020-01-08T21:39:16Z

sansa-ml-spark/src/main/scala/net/sansa_stack/ml/spark/entity_resolution/Commons.scala

@@ -71,16 +75,19 @@ class EntityResolution_RDFData_CountVectorizer(spark: SparkSession, triplesRDD1:
   * where subject is the key and the group of paired predicate:object forms the value
   * ex:- (Budapest_City, areaCode:1 , country:Hungary)
   */
-  def get_entity_profiles(triplesRDD: RDD[Triple]): RDD[(String, String)] = {
+  def getEntityProfiles(sourceData: RDD[Triple]): RDD[(String, String)] = {

    //predicates to be filtered out from triples


Not, sure if you have run mvn scalastyle:check before you pushed the changes, but what I can see Scala Style Check will complain about this one (see here: http://www.scalastyle.org/rules-dev.html#org_scalastyle_scalariform_SpaceAfterCommentStartChecker ).

Do the same for the rest!

sansa-ml-spark/src/main/scala/net/sansa_stack/ml/spark/entity_resolution/Commons.scala

...ml-spark/src/main/scala/net/sansa_stack/ml/spark/entity_resolution/predicatesfilteration.txt

GezimSejdiu · 2020-02-07T22:02:51Z

Hi @amrit1210 ,

any news about my comments? It would be good if we can resolve some of them and move the PR forward.

Many thanks in advance for your valuable contribution.

Best regards,

…b.com/amrit1210/SANSA-ML into entity_resolution/feature_integration

amrit1210 · 2020-02-12T23:27:04Z

Dear Gezim, The suggested changes have been made and updated to the git. Please review. Thanks and Regards, Amrit Kaur

…

On Fri, Feb 7, 2020 at 11:02 PM Gezim Sejdiu ***@***.***> wrote: Hi @amrit1210 <https://github.com/amrit1210> , any news about my comments? It would be good if we can resolve some of them and move the PR forward. Many thanks in advance for your valuable contribution. Best regards, — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#24>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGVISQHDDEAVMRHODLGU5ILRBXLAXANCNFSM4JFSM6FA> .

amrit1210 · 2020-02-13T15:48:59Z

Dear Gezim,

To the last patch, I have added some more corrections. Please consider both the commits for complete changes.

Kind Regards,

GezimSejdiu · 2020-02-19T22:17:05Z

Hi @amrit1210 ,

once more, thanks a lot for your contribution.
I just run the PR locally and it seems to fail in one of the test-cases. I'm getting:

ERTests:
net.sansa_stack.ml.spark.entity_resolution.ERTests *** ABORTED ***
  java.lang.RuntimeException: Unable to load a Suite class that was discovered in the runpath: net.sansa_stack.ml.spark.entity_resolution.ERTests
  at org.scalatest.tools.DiscoverySuite$.getSuiteInstance(DiscoverySuite.scala:81)
  at org.scalatest.tools.DiscoverySuite$$anonfun$1.apply(DiscoverySuite.scala:38)
  at org.scalatest.tools.DiscoverySuite$$anonfun$1.apply(DiscoverySuite.scala:37)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234)
  at scala.collection.Iterator$class.foreach(Iterator.scala:891)
  at scala.collection.AbstractIterator.foreach(Iterator.scala:1334)
  at scala.collection.IterableLike$class.foreach(IterableLike.scala:72)
  at scala.collection.AbstractIterable.foreach(Iterable.scala:54)
  at scala.collection.TraversableLike$class.map(TraversableLike.scala:234)
  ...
  Cause: java.lang.NullPointerException:
  at net.sansa_stack.rdf.spark.io.NTripleReader$.load(NTripleReader.scala:112)
  at net.sansa_stack.rdf.spark.io.package$RDFReader$$anonfun$ntriples$4.apply(package.scala:217)
  at net.sansa_stack.rdf.spark.io.package$RDFReader$$anonfun$ntriples$4.apply(package.scala:216)
  at net.sansa_stack.ml.spark.entity_resolution.ERTests.<init>(ERTests.scala:17)
  at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
  at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62)
  at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45)
  at java.lang.reflect.Constructor.newInstance(Constructor.java:423)
  at java.lang.Class.newInstance(Class.java:442)
  at org.scalatest.tools.DiscoverySuite$.getSuiteInstance(DiscoverySuite.scala:66)
  ...

Would it be possible that you double-check that test case(s) via mvn test and see what is wrong with that?

Best regards,

amrit1210 · 2020-02-19T23:05:38Z

Hi Gezim, I have rechecked. Unfortunately, the sparkSession here is null. I am not sure why. Can you please give more insight into it?. Regards, Amrit Kaur

…

On Wed, Feb 19, 2020, 11:17 PM Gezim Sejdiu ***@***.***> wrote: Hi @amrit1210 <https://github.com/amrit1210> , once more, thanks a lot for your contribution. I just run the PR locally and it seems to fail in one of the test-cases. I'm getting: ERTests: net.sansa_stack.ml.spark.entity_resolution.ERTests *** ABORTED *** java.lang.RuntimeException: Unable to load a Suite class that was discovered in the runpath: net.sansa_stack.ml.spark.entity_resolution.ERTests at org.scalatest.tools.DiscoverySuite$.getSuiteInstance(DiscoverySuite.scala:81) at org.scalatest.tools.DiscoverySuite$$anonfun$1.apply(DiscoverySuite.scala:38) at org.scalatest.tools.DiscoverySuite$$anonfun$1.apply(DiscoverySuite.scala:37) at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234) at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234) at scala.collection.Iterator$class.foreach(Iterator.scala:891) at scala.collection.AbstractIterator.foreach(Iterator.scala:1334) at scala.collection.IterableLike$class.foreach(IterableLike.scala:72) at scala.collection.AbstractIterable.foreach(Iterable.scala:54) at scala.collection.TraversableLike$class.map(TraversableLike.scala:234) ... Cause: java.lang.NullPointerException: at net.sansa_stack.rdf.spark.io.NTripleReader$.load(NTripleReader.scala:112) at net.sansa_stack.rdf.spark.io.package$RDFReader$$anonfun$ntriples$4.apply(package.scala:217) at net.sansa_stack.rdf.spark.io.package$RDFReader$$anonfun$ntriples$4.apply(package.scala:216) at net.sansa_stack.ml.spark.entity_resolution.ERTests.<init>(ERTests.scala:17) at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method) at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62) at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45) at java.lang.reflect.Constructor.newInstance(Constructor.java:423) at java.lang.Class.newInstance(Class.java:442) at org.scalatest.tools.DiscoverySuite$.getSuiteInstance(DiscoverySuite.scala:66) ... Would it be possible that you double-check that test case(s) via mvn test and see what is wrong with that? Best regards, — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#24>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGVISQESMRPP6HHIRPYQ543RDWVWFANCNFSM4JFSM6FA> .

amrit1210 · 2020-02-29T17:58:48Z

Hi Gezim,

I have rechecked it again. But I am not sure about the null session. Can you please check that?.

Thanks and Regards,
Amrit Kaur

…eption)

GezimSejdiu · 2020-03-20T23:28:01Z

Hi @amrit1210 ,
so I just did a clean up and fixed the issue within the null value to the SparkSession but it is throwing another exception w.r.t to the datatypes. See the trace below:

20/03/21 00:25:51 ERROR Executor: Exception in task 16.0 in stage 3.0 (TID 76)
org.apache.jena.datatypes.DatatypeFormatException: Lexical form '1967-4-4' is not a legal instance of Datatype[http://www.w3.org/2001/XMLSchema#date] Lexical form '1967-4-4' is not a legal instance of Datatype[http://www.w3.org/2001/XMLSchema#date] during parse -org.apache.jena.ext.xerces.impl.dv.InvalidDatatypeValueException: cvc-datatype-valid.1.2.1: '1967-4-4' is not a valid value for 'date'.
	at org.apache.jena.graph.impl.LiteralLabelImpl.getValue(LiteralLabelImpl.java:338)
	at org.apache.jena.graph.Node_Literal.getLiteralValue(Node_Literal.java:44)
	at net.sansa_stack.ml.spark.entity_resolution.Commons$$anonfun$3.apply(Commons.scala:91)
	at net.sansa_stack.ml.spark.entity_resolution.Commons$$anonfun$3.apply(Commons.scala:83)
	at scala.collection.Iterator$$anon$11.next(Iterator.scala:410)
	at org.apache.spark.util.collection.ExternalSorter.insertAll(ExternalSorter.scala:193)
	at org.apache.spark.shuffle.sort.SortShuffleWriter.write(SortShuffleWriter.scala:62)
	at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:99)
	at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:55)
	at org.apache.spark.scheduler.Task.run(Task.scala:123)
	at org.apache.spark.executor.Executor$TaskRunner$$anonfun$10.apply(Executor.scala:408)
	at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1360)
	at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:414)
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
	at java.lang.Thread.run(Thread.java:748)
20/03/21 00:25:52 ERROR TaskSetManager: Task 16 in stage 3.0 failed 1 times; aborting job

Please, could you double check if the data you are giving as an input (if you have tested them before) are valid or we have to chnge something on the transformation (i.e. when you are getting the datatypes of the literals).

Best regards,

amrit1210 · 2020-03-28T21:27:30Z

Hi Gezim,

The issue is fixed. Tested.

Regards,
Amrit

GezimSejdiu · 2020-04-03T20:41:40Z

sansa-ml-spark/src/main/scala/net/sansa_stack/ml/spark/entity_resolution/Commons.scala

@@ -88,7 +88,7 @@ abstract class Commons(val spark: SparkSession, val sourceData1: RDD[Triple], va
          val value = pred + ":" + obj // predicate and object are seperated by ':'
          (key, value)
        } else {
-          val obj = f.getObject.getLiteralValue.toString()
+          val obj = f.getObject.getLiteral.toString().split(Array('^', '@')).head.trim()


Would it be better if we get the lexical form of the literal? which will remove everything i.e. language tag, type, etc? Or is this method meant for something else?

Yes, It does the same.

LorenzBuehmann

I would not accept this PR in its current state (but maybe it's sufficient enough for you and the project? I wasn't involved in this algorithm so I don't know the requirements you defined, I just know that "working on DBpedia dataset" doesn't mean anything in general)

for me the whole purpose of the algorithm is basically a blackbox. I neither know what input nor what output is supposed to be, so it's hard to judge on whether it's doing what it is supposed to do
there is a bunch of params that are not documented, so it's hard to understand their purpose, like having two threshold params for subject and object or two config params for number of partitions and repartition

The most critical point:

The whole concept of using URIs as identifiers is ignored by just keeping the local name during runtime. This might be ok at a first glance, but a property with the same local name but different namespaces can have a totally different meaning.
Moreover, the result of the whole pipeline is also just a tuple of two strings which are just the local names, so you won't have a direct mapping back to the URIs of the original data. This again is error prone.
Also, the blacklist of predicates considered during computation is basically hard coded for DBpedia.
there are some (in my opinion) inefficient parts in the code like doing (pseudo code)

triplesToRemove = triples.filter(t.p => p in blacklist)
triplesToKeep = triples.subtract(triplesToRemove)

instead of just doing negation on the filter, e.g.

triplesToKeep = triples.filter(t.p => p **not** in blacklist)

is it really necessary to use string operations back and forth? like e = "p:o" and later you do e.split(":") to get p and/or o. wouldn't it be easier to keep tuples here? Or do you need the string "p:o" for something important?
many Scaladoc params do not match the method signature (anymore)

Adding entity resolution feature to SANSA-ML. The feature uses minhas…

f323e7b

…hLSH based approach to identify and link similar entities in two datasets

GezimSejdiu self-requested a review November 17, 2019 20:32

GezimSejdiu reviewed Nov 17, 2019

View reviewed changes

updated changes as per review

684cc6d

GezimSejdiu reviewed Jan 8, 2020

View reviewed changes

Merge branch 'develop' into entity_resolution/feature_integration

26c9e5c

GezimSejdiu added the ML label Jan 15, 2020

GezimSejdiu added this to the 0.8 milestone Jan 15, 2020

Merge branch 'develop' into entity_resolution/feature_integration

5b876d3

amrit1210 added 2 commits February 13, 2020 00:15

Suggested changes made

5e5aace

Merge branch 'entity_resolution/feature_integration' of https://githu…

24caf70

…b.com/amrit1210/SANSA-ML into entity_resolution/feature_integration

Final changes

4f9c7fe

Make ERTests suite works (still issue with the data DatatypeFormatExc…

6e92b0d

…eption)

Bugfix

70ce1b7

GezimSejdiu reviewed Apr 3, 2020

View reviewed changes

Bugfix

ee7120c

GezimSejdiu approved these changes Apr 25, 2020

View reviewed changes

GezimSejdiu requested review from carstendraschner, HajiraJabeen and LorenzBuehmann May 2, 2020 21:16

LorenzBuehmann suggested changes May 4, 2020

View reviewed changes

JensLehmann removed this from the 0.8 milestone Jul 2, 2020

JensLehmann added this to the 0.9 milestone Jul 2, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Adding entity resolution feature to SANSA-ML. #24

Adding entity resolution feature to SANSA-ML. #24

amrit1210 commented Oct 27, 2019

GezimSejdiu left a comment

GezimSejdiu Nov 17, 2019

GezimSejdiu Nov 17, 2019

GezimSejdiu Nov 17, 2019

GezimSejdiu Nov 17, 2019

GezimSejdiu Nov 17, 2019

GezimSejdiu Nov 17, 2019

GezimSejdiu Nov 17, 2019

GezimSejdiu Nov 17, 2019

GezimSejdiu left a comment

GezimSejdiu Jan 8, 2020

amrit1210 Jan 11, 2020

GezimSejdiu Jan 12, 2020

GezimSejdiu Jan 8, 2020

amrit1210 Jan 11, 2020

GezimSejdiu Jan 12, 2020

GezimSejdiu Jan 8, 2020

amrit1210 Feb 10, 2020

GezimSejdiu Jan 8, 2020

GezimSejdiu Jan 8, 2020

amrit1210 Feb 11, 2020

GezimSejdiu commented Feb 7, 2020

amrit1210 commented Feb 12, 2020 via email

amrit1210 commented Feb 13, 2020

GezimSejdiu commented Feb 19, 2020

amrit1210 commented Feb 19, 2020 via email •

edited

Loading

amrit1210 commented Feb 29, 2020

GezimSejdiu commented Mar 20, 2020

amrit1210 commented Mar 28, 2020

GezimSejdiu Apr 3, 2020

amrit1210 Apr 4, 2020

LorenzBuehmann left a comment •

edited

Loading

		@@ -35,22 +39,22 @@ class EntityResolution_RDFData_CountVectorizer(spark: SparkSession, triplesRDD1:
		def run(): RDD[(String, String)] = {

Adding entity resolution feature to SANSA-ML. #24

Are you sure you want to change the base?

Adding entity resolution feature to SANSA-ML. #24

Conversation

amrit1210 commented Oct 27, 2019

GezimSejdiu left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

GezimSejdiu left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

GezimSejdiu commented Feb 7, 2020

amrit1210 commented Feb 12, 2020 via email

amrit1210 commented Feb 13, 2020

GezimSejdiu commented Feb 19, 2020

amrit1210 commented Feb 19, 2020 via email • edited Loading

amrit1210 commented Feb 29, 2020

GezimSejdiu commented Mar 20, 2020

amrit1210 commented Mar 28, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

LorenzBuehmann left a comment • edited Loading

Choose a reason for hiding this comment

amrit1210 commented Feb 19, 2020 via email •

edited

Loading

LorenzBuehmann left a comment •

edited

Loading