Matrix Factorization XML docs #3409

wschin · 2019-04-18T20:42:35Z

Toward #2522 following #3218.

codecov · 2019-04-18T21:20:51Z

Codecov Report

Merging #3409 into master will increase coverage by 0.04%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #3409      +/-   ##
==========================================
+ Coverage   72.69%   72.74%   +0.04%     
==========================================
  Files         807      807              
  Lines      145171   145206      +35     
  Branches    16225    16230       +5     
==========================================
+ Hits       105536   105632      +96     
+ Misses      35220    35153      -67     
- Partials     4415     4421       +6

Flag	Coverage Δ
#Debug	`72.74% <ø> (+0.04%)`	⬆️
#production	`68.29% <ø> (+0.06%)`	⬆️
#test	`88.98% <ø> (ø)`	⬆️

Impacted Files	Coverage Δ
...osoft.ML.Recommender/MatrixFactorizationTrainer.cs	`70.39% <ø> (ø)`	⬆️
src/Microsoft.ML.Recommender/RecommenderCatalog.cs	`70.83% <ø> (ø)`	⬆️
src/Microsoft.ML.Data/Transforms/KeyToValue.cs	`79.16% <0%> (-0.65%)`	⬇️
src/Microsoft.ML.Transforms/Text/LdaTransform.cs	`89.26% <0%> (-0.63%)`	⬇️
...ML.Transforms/MutualInformationFeatureSelection.cs	`78.58% <0%> (-0.33%)`	⬇️
...ML.Data/Transforms/ColumnConcatenatingEstimator.cs	`80.3% <0%> (ø)`	⬆️
...soft.ML.Transforms/Text/WordEmbeddingsExtractor.cs	`87.52% <0%> (ø)`	⬆️
...icrosoft.ML.FastTree/RandomForestClassification.cs	`78.07% <0%> (ø)`	⬆️
...forms/FeatureContributionCalculationTransformer.cs	`73.55% <0%> (ø)`	⬆️
...soft.ML.Tests/Transformers/CategoricalHashTests.cs	`100% <0%> (ø)`	⬆️
... and 42 more

singlis · 2019-04-18T22:44:43Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// or [MatrixFactorization(Options)](xref:Microsoft.ML.RecommendationCatalog.RecommendationTrainers.MatrixFactorization(Microsoft.ML.Trainers.MatrixFactorizationTrainer.Options)).
+    ///
+    /// ### Input and Output Columns
+    /// There are three input columns required, one for matrix row indexes, one for matrix column indexes, and one for values (i.e., labels) in matrix .


[](start = 150, length = 1)

extra space. #Resolved

singlis · 2019-04-18T22:44:58Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    ///
+    /// ### Input and Output Columns
+    /// There are three input columns required, one for matrix row indexes, one for matrix column indexes, and one for values (i.e., labels) in matrix .
+    /// They together defines a matrix in [COO](https://en.wikipedia.org/wiki/Sparse_matrix#Coordinate_list_(COO)) format.


defines [](start = 22, length = 7)

define a matrix #Resolved

singlis · 2019-04-18T22:46:34Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// ### Input and Output Columns
+    /// There are three input columns required, one for matrix row indexes, one for matrix column indexes, and one for values (i.e., labels) in matrix .
+    /// They together defines a matrix in [COO](https://en.wikipedia.org/wiki/Sparse_matrix#Coordinate_list_(COO)) format.
+    /// The type of label column is vector of <xref:System.Single> while the other two columns are key-typed scalar.


The type of label column is vector of xref:System.Single [](start = 8, length = 58)

suggested: The type for label column is a vector of xref:System.Single ... #Resolved

singlis · 2019-04-18T22:47:40Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// ### Input and Output Columns
+    /// There are three input columns required, one for matrix row indexes, one for matrix column indexes, and one for values (i.e., labels) in matrix .
+    /// They together defines a matrix in [COO](https://en.wikipedia.org/wiki/Sparse_matrix#Coordinate_list_(COO)) format.
+    /// The type of label column is vector of <xref:System.Single> while the other two columns are key-typed scalar.


key-typed scalar [](start = 99, length = 16)

Key-typed scalar? What does that mean? Are you talking about KeyType? Should this have an xref? #Resolved

Will do [key-typed](<xref:Microsoft.ML.Data.KeyDataViewType>) scalar

In reply to: 276858813 [](ancestors = 276858813)

singlis · 2019-04-18T22:48:59Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// | Required NuGet in addition to Microsoft.ML | Microsoft.ML.Recommender |
+    ///
+    /// ### Background
+    /// The basic idea of matrix factorization is finding two low-rank factor marcies to apporimate the training matrix.


marcies [](start = 78, length = 7)

matrices? #Resolved

Yes!

In reply to: 276859081 [](ancestors = 276859081)

singlis · 2019-04-18T22:49:12Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// | Required NuGet in addition to Microsoft.ML | Microsoft.ML.Recommender |
+    ///
+    /// ### Background
+    /// The basic idea of matrix factorization is finding two low-rank factor marcies to apporimate the training matrix.


apporimate [](start = 89, length = 10)

approximate? #Resolved

singlis · 2019-04-18T22:50:10Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    ///
+    /// ### Background
+    /// The basic idea of matrix factorization is finding two low-rank factor marcies to apporimate the training matrix.
+    /// In this module, the expected training data, the factorized matrix, is a list of tuples. Every tuple consists of a column index, a row index,


, [](start = 50, length = 1)

suggestion: the expected training data and the factorized matrix is a list of tuples. #Resolved

Will do the expected training data (the factorized matrix) is a list of tuples because training data = factorized matrix.

In reply to: 276859310 [](ancestors = 276859310)

singlis · 2019-04-18T22:50:39Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    ///
+    /// ### Background
+    /// The basic idea of matrix factorization is finding two low-rank factor marcies to apporimate the training matrix.
+    /// In this module, the expected training data, the factorized matrix, is a list of tuples. Every tuple consists of a column index, a row index,
    /// and the value at the location specified by the two indexes. For an example data structure of a tuple, one can use:


indexes [](start = 59, length = 7)

indices? #Resolved

Ok.

In reply to: 276859401 [](ancestors = 276859401)

singlis · 2019-04-18T22:57:53Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

-    /// <![CDATA[
-    /// [!code-csharp[MF](~/../docs/samples/docs/samples/Microsoft.ML.Samples/Dynamic/Trainers/Recommendation/MatrixFactorization.cs)]
+    /// Assume that $R\in{\mathbb R}^{m\times n}$ is a m-by-n rating matrix and the [rank](https://en.wikipedia.org/wiki/Rank_(linear_algebra)) of the two factor matrices are $P\in {\mathbb R}^{k\times m}$ and $Q\in {\mathbb R}^{k\times n}$, where $k$ is the approximation rank.
+    /// The predicted rating at the $u$-th row and the $v$-th column in $R$ would be the inner product of the $u$-th row of P and the $v$-th row of Q; that is, $R$ is approximated by the product of $P$'s transpose ($P^T$) and $Q$.


$R$ [](start = 72, length = 3)

R is referenced in latex, but P and Q are not. #Resolved

Nice catch!

In reply to: 276860792 [](ancestors = 276860792)

singlis · 2019-04-18T22:58:34Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// The predicted rating at the $u$-th row and the $v$-th column in $R$ would be the inner product of the $u$-th row of P and the $v$-th row of Q; that is, $R$ is approximated by the product of $P$'s transpose ($P^T$) and $Q$.
+    /// Note that $k$ is usually much smaller than $m$ and $n$, so $P^T Q$ is usually called a low-rank approximation of $R$.
+    ///
+    /// This trainer includes a [stochastic gradient method](https://en.wikipedia.org/wiki/Stochastic_gradient_descent) and [a coordinate descent method](https://en.wikipedia.org/wiki/Coordinate_descent) for finding $P$ and $Q$ via minimizing the distance between (non-missing part of) $R$ and its approximation $P^T Q$.


a [](start = 125, length = 1)

minor, should 'a' be part of the uri? #Resolved

It shouldn't. CD is a family of algorithms.

In reply to: 276860936 [](ancestors = 276860936)

singlis · 2019-04-18T23:00:35Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    ///
+    /// This trainer includes a [stochastic gradient method](https://en.wikipedia.org/wiki/Stochastic_gradient_descent) and [a coordinate descent method](https://en.wikipedia.org/wiki/Coordinate_descent) for finding $P$ and $Q$ via minimizing the distance between (non-missing part of) $R$ and its approximation $P^T Q$.
+    /// The coordinate descent method included is specifically for one-class matrix factorization where all observed ratings are positive signals (that is, all rating values are 1).
+    /// Notice taht the only way to invoke one-class matrix factorization is to assign <see cref="Microsoft.ML.Trainers.MatrixFactorizationTrainer.LossFunctionType.SquareLossOneClass"/> to [loss function](Microsoft.ML.Trainers.MatrixFactorizationTrainer.Options.LossFunction)


cref [](start = 92, length = 4)

cref in markdown?? who does that? :) #Resolved

singlis · 2019-04-18T23:00:42Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    ///
+    /// This trainer includes a [stochastic gradient method](https://en.wikipedia.org/wiki/Stochastic_gradient_descent) and [a coordinate descent method](https://en.wikipedia.org/wiki/Coordinate_descent) for finding $P$ and $Q$ via minimizing the distance between (non-missing part of) $R$ and its approximation $P^T Q$.
+    /// The coordinate descent method included is specifically for one-class matrix factorization where all observed ratings are positive signals (that is, all rating values are 1).
+    /// Notice taht the only way to invoke one-class matrix factorization is to assign <see cref="Microsoft.ML.Trainers.MatrixFactorizationTrainer.LossFunctionType.SquareLossOneClass"/> to [loss function](Microsoft.ML.Trainers.MatrixFactorizationTrainer.Options.LossFunction)


taht [](start = 15, length = 4)

that #Resolved

singlis · 2019-04-18T23:01:23Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// Notice taht the only way to invoke one-class matrix factorization is to assign <see cref="Microsoft.ML.Trainers.MatrixFactorizationTrainer.LossFunctionType.SquareLossOneClass"/> to [loss function](Microsoft.ML.Trainers.MatrixFactorizationTrainer.Options.LossFunction)
+    /// when calling [MatrixFactorization(Options)](xref:Microsoft.ML.RecommendationCatalog.RecommendationTrainers.MatrixFactorization(Microsoft.ML.Trainers.MatrixFactorizationTrainer.Options)).
+    /// See Page 28 [here](https://www.csie.ntu.edu.tw/~cjlin/talks/facebook.pdf) for a brief introduction to one-class matrix factorization.
+    /// The default settting <see cref="MatrixFactorizationTrainer.LossFunctionType.SquareLossRegression"/> is for standard matrix factorization problem.


cref [](start = 34, length = 4)

another cref #Resolved

singlis · 2019-04-18T23:02:38Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// For users interested in the mathematical details, please see the references below.
+    ///
+    /// * For the multi-threading implementation of the used stochastic gradient method, see [A Fast Parallel Stochastic Gradient Method for Matrix Factorization in Shared Memory Systems](https://www.csie.ntu.edu.tw/~cjlin/papers/libmf/libmf_journal.pdf).
+    /// * For the computation happened inside one thread, see [A Learning-rate Schedule for Stochastic Gradient Methods to Matrix Factorization](https://www.csie.ntu.edu.tw/~cjlin/papers/libmf/mf_adaptive_pakdd.pdf).


happened [](start = 30, length = 8)

happening inside a single thread? #Resolved

singlis · 2019-04-18T23:03:50Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

+    /// </remarks>
+    /// <seealso cref="Microsoft.ML.RecommendationCatalog.RecommendationTrainers.MatrixFactorization(System.String,System.String,System.String,System.Int32,System.Double,System.Int32)"/>
+    /// <seealso cref="Microsoft.ML.RecommendationCatalog.RecommendationTrainers.MatrixFactorization(Microsoft.ML.Trainers.MatrixFactorizationTrainer.Options)"/>
+    /// <seealso cref="Options"/>


Options [](start = 23, length = 7)

is this one needed? #Resolved

yes

In reply to: 276861871 [](ancestors = 276861871)

singlis · 2019-04-18T23:07:14Z

src/Microsoft.ML.Recommender/RecommenderCatalog.cs

@@ -80,6 +80,8 @@ internal RecommendationTrainers(RecommendationCatalog catalog)
            /// <para>The basic idea of matrix factorization is finding two low-rank factor matrices to apporimate the training matrix.</para>


apporimate [](start = 104, length = 10)

approximate #Resolved

please add spell-checker plugin, if you haven't already
https://marketplace.visualstudio.com/items?itemName=EWoodruff.VisualStudioSpellChecker

In reply to: 276862518 [](ancestors = 276862518)

Super helpful tip. Thanks.

In reply to: 276869057 [](ancestors = 276869057,276862518)

singlis

shmoradims · 2019-04-18T23:52:46Z

src/Microsoft.ML.Recommender/RecommenderCatalog.cs

@@ -80,6 +80,8 @@ internal RecommendationTrainers(RecommendationCatalog catalog)
            /// <para>The basic idea of matrix factorization is finding two low-rank factor matrices to apporimate the training matrix.</para>
            /// <para>In this module, the expected training data is a list of tuples. Every tuple consists of a column index, a row index,
            /// and the value at the location specified by the two indexes. The training configuration is encoded in <see cref="MatrixFactorizationTrainer.Options"/>.
+            /// To invoke one-class matrix factorization, user needs to specify <see cref="MatrixFactorizationTrainer.LossFunctionType.SquareLossOneClass"/>.
+            /// The default settting <see cref="MatrixFactorizationTrainer.LossFunctionType.SquareLossRegression"/> is for standard matrix factorization problem.


we're moving all the algo-details remarks to the estimator class. you can just link there instead of repeating the content. #ByDesign

Maybe no because 1) this thing I have explained to an user 3 times on Gitter 2) this is the switch of implicit-feedback and explicit-feed recommendation --- a formula change rather than an algorithm change.

In reply to: 276869690 [](ancestors = 276869690)

shmoradims · 2019-04-18T23:53:10Z

src/Microsoft.ML.Recommender/RecommenderCatalog.cs

@@ -80,6 +80,8 @@ internal RecommendationTrainers(RecommendationCatalog catalog)
            /// <para>The basic idea of matrix factorization is finding two low-rank factor matrices to apporimate the training matrix.</para>
            /// <para>In this module, the expected training data is a list of tuples. Every tuple consists of a column index, a row index,
            /// and the value at the location specified by the two indexes. The training configuration is encoded in <see cref="MatrixFactorizationTrainer.Options"/>.
+            /// To invoke one-class matrix factorization, user needs to specify <see cref="MatrixFactorizationTrainer.LossFunctionType.SquareLossOneClass"/>.
+            /// The default settting <see cref="MatrixFactorizationTrainer.LossFunctionType.SquareLossRegression"/> is for standard matrix factorization problem.
            /// </para>
            /// </remarks>
            /// <param name="options">Advanced arguments to the algorithm.</param>


Advanced arguments to the algorithm. [](start = 38, length = 36)

"Trainer options." is the current template. #Resolved

shmoradims · 2019-04-18T23:53:31Z

        /// Train a matrix factorization model. It factorizes the training matrix into the product of two low-rank matrices.

please follow template #Resolved

Refers to: src/Microsoft.ML.Recommender/RecommenderCatalog.cs:77 in 000aa46. [](commit_id = 000aa46, deletion_comment = False)

shmoradims · 2019-04-18T23:53:51Z

        /// Train a matrix factorization model. It factorizes the training matrix into the product of two low-rank matrices.

same here #Resolved

Refers to: src/Microsoft.ML.Recommender/RecommenderCatalog.cs:46 in 000aa46. [](commit_id = 000aa46, deletion_comment = False)

shmoradims · 2019-04-18T23:54:17Z

        /// <param name="matrixColumnIndexColumnName">The name of the column hosting the matrix's column IDs.</param>

is there any data type information here? #Resolved

Refers to: src/Microsoft.ML.Recommender/RecommenderCatalog.cs:55 in 000aa46. [](commit_id = 000aa46, deletion_comment = False)

shmoradims · 2019-04-18T23:55:37Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

@@ -26,11 +26,33 @@ namespace Microsoft.ML.Trainers
    /// Train a matrix factorization model. It factorizes the training matrix into the product of two low-rank matrices.


/// Train a matr [](start = 4, length = 16)

please follow template #Resolved

shmoradims · 2019-04-18T23:59:37Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

    /// ]]>
    /// </format>
-    /// </example>
+    /// </remarks>
+    /// <seealso cref="Microsoft.ML.RecommendationCatalog.RecommendationTrainers.MatrixFactorization(System.String,System.String,System.String,System.Int32,System.Double,System.Int32)"/>


System.String [](start = 101, length = 13)

no need to change these to fully qualified names. cref is more forgiving. whatever link visual studio intellisense gives you should be good. #Resolved

Just want to reuse my xref. I could easily make mistake when modifying those things.

In reply to: 276870656 [](ancestors = 276870656)

intellisense will do the work for you. I'm not even certain xref-UID would work with cref.

please change as we discussed on Teams.

In reply to: 277024650 [](ancestors = 277024650,276870656)

shmoradims · 2019-04-19T23:44:03Z

src/Microsoft.ML.Recommender/MatrixFactorizationTrainer.cs

@@ -109,22 +134,25 @@ public enum LossFunctionType
        };

        /// <summary>
-        /// Advanced options for the <see cref="MatrixFactorizationTrainer"/>.
+        /// Options for the <see cref="MatrixFactorizationTrainer"/> in [MatrixFactorization(Options)](xref:Microsoft.ML.RecommendationCatalog.RecommendationTrainers.MatrixFactorization(Microsoft.ML.Trainers.MatrixFactorizationTrainer.Options)).


in [](start = 69, length = 2)

as used in #Resolved

shmoradims

MF XML

000aa46

wschin added the documentation Related to documentation of ML.NET label Apr 18, 2019

wschin requested review from Ivanidzo4ka and sfilipi April 18, 2019 20:42

wschin self-assigned this Apr 18, 2019

shmoradims mentioned this pull request Apr 18, 2019

Docs and samples for the API reference site (P0 & P1 Trainers) #2522

Closed

singlis reviewed Apr 18, 2019

View reviewed changes

singlis approved these changes Apr 18, 2019

View reviewed changes

shmoradims reviewed Apr 18, 2019

View reviewed changes

wschin requested a review from natke April 19, 2019 15:23

Address comments

d1f8fb9

wschin requested a review from shmoradims April 19, 2019 16:30

Adress comments

22a9a44

shmoradims reviewed Apr 19, 2019

View reviewed changes

shmoradims approved these changes Apr 19, 2019

View reviewed changes

wschin added 2 commits April 19, 2019 17:31

Fix cref

6ed06bc

Comments

33b2a29

wschin merged commit 3fa207d into dotnet:master Apr 20, 2019

wschin deleted the xml-mf branch April 20, 2019 03:55

ghost locked as resolved and limited conversation to collaborators Mar 22, 2022

		@@ -80,6 +80,8 @@ internal RecommendationTrainers(RecommendationCatalog catalog)
		/// <para>The basic idea of matrix factorization is finding two low-rank factor matrices to apporimate the training matrix.</para>

		@@ -26,11 +26,33 @@ namespace Microsoft.ML.Trainers
		/// Train a matrix factorization model. It factorizes the training matrix into the product of two low-rank matrices.

Matrix Factorization XML docs #3409

Matrix Factorization XML docs #3409

Conversation

wschin commented Apr 18, 2019

codecov bot commented Apr 18, 2019 • edited Loading

Codecov Report

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

wschin Apr 19, 2019 • edited Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis left a comment

Choose a reason for hiding this comment

shmoradims Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

shmoradims Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

shmoradims commented Apr 18, 2019 • edited by wschin Loading

shmoradims commented Apr 18, 2019 • edited by wschin Loading

shmoradims commented Apr 18, 2019 • edited by wschin Loading

shmoradims Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

shmoradims Apr 18, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

shmoradims Apr 19, 2019 • edited by wschin Loading

Choose a reason for hiding this comment

shmoradims left a comment

Choose a reason for hiding this comment

codecov bot commented Apr 18, 2019 •

edited

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

wschin Apr 19, 2019 •

edited

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

singlis Apr 18, 2019 •

edited by wschin

Loading

shmoradims Apr 18, 2019 •

edited by wschin

Loading

shmoradims Apr 18, 2019 •

edited by wschin

Loading

shmoradims commented Apr 18, 2019 •

edited by wschin

Loading

shmoradims commented Apr 18, 2019 •

edited by wschin

Loading

shmoradims commented Apr 18, 2019 •

edited by wschin

Loading

shmoradims Apr 18, 2019 •

edited by wschin

Loading

shmoradims Apr 18, 2019 •

edited by wschin

Loading

shmoradims Apr 19, 2019 •

edited by wschin

Loading