Schema based text loader #1878

sfilipi · 2018-12-14T00:23:35Z

Initial work to address #561.

It moves the current reflection-based logic outside of the legacy project, into the TextLoader catalog, creating an API for it.
Decouples the LoadColumn and Column attributes.

Ivanidzo4ka · 2018-12-14T18:47:13Z

test/Microsoft.ML.Tests/ScenariosWithDirectInstantiation/TensorflowTests.cs

            public long Label;

-            [Column(ordinal: "1-784")]
+            [LoadColumn(range: "1-784")]


LoadColumn [](start = 13, length = 10)

I think we need some good word smith here.
It's probably slightly better than just column, but still confusing, especially next to that ColumnName attribute few lines below.

Can it be TextLoader.MarkingColumn? TextLoader.MappingColumn, something related to textloader since it's used only for Textloader (right?)
#Pending

It is called LoadColumn, if you are referring to the new attribute.

In reply to: 241852931 [](ancestors = 241852931)

wschin · 2018-12-14T20:23:08Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoaderSaverCatalog.cs

+            bool supportSparse = TextLoader.DefaultArguments.AllowSparse,
+            bool trimWhitespace = TextLoader.DefaultArguments.TrimWhitespace)
+        {
+            var userType = typeof(TInput);


Can you describe what this function is doing? I'd like to see something ComputeOutputSchema in this PR where high-level description is added to header and low-level descriptions are inline to explain high-level concept in detail. #Resolved

Introducing test.

sfilipi · 2018-12-18T07:06:10Z

Reading the description of both #561 and #1515, i prototyped around having more attribute parameters for the LoadColumn attribute, to get closed to the TextLoader.Range.
In the process, discovered that the members for the Attribute derived class cannot be of the more advanced types: like classes, nullable primitives (ints), tuples.

I tried to eliminate the need for the existing Ordinal string by adding a start/end string (can't do int? ), int[] columns and int[]inverseColumns thinking they would suffice to more strongly type and support most of the scenarios. And with them we can do the
1-* - auto end
1-4 - defined ranges
1, 2, 3 - defined columns, or inverse columns

but the scenario that we can't do is multiple ranges. i.e. 1-4, 5-6

One idea on how to convert that would be to provide an array where the users would specify the indexes with the convention that they should be treated in groups; example: new [] { 1, 3, 5, 7} would translate to 1-3, 5-7 but that seems UGLY.

Another idea is to let the user specify multiple attributes for the range, and have the column name specified separately. Example:

 [LoadColumn(start:"0", end: "8")]
 [LoadColumn(start:"10", end: "18")]
 [ColumnName( "Features")]
 public float SepalLength;

But, IMHO, the range regex beats this too from the prospective of simplicity and ease of use.

So to summarize, i propose:
1- We keep the old ordinal, as string, rename to range
2- We add a boolean LoadInverseRange to determine whether we add the columns defined in range or they should be the one not loaded (AllOther in TextLoader.Range)

The construction of the TextLoader columns from the attributes is not final. I will complete after getting a final word here.

@eerhardt, @TomFinley, @Zruty0, @CESARDELATORRE #Resolved

sfilipi · 2018-12-18T07:07:06Z

Reading the description of both #561 and #1515, i prototyped around having more attribute parameters for the LoadColumn attribute, to get closed to the TextLoader.Range.
In the process, discovered that the members for the Attribute derived class cannot be of the more advanced types: like classes, nullable primitives (ints), tuples.

I tried to eliminate the need for the existing Ordinal string by adding a start/end string (can't do int? ), int[] columns and int[]inverseColumns thinking they would suffice to more strongly type and support most of the scenarios. And with them we can do the
1-* - auto end
1-4 - defined ranges
1, 2, 3 - defined columns, or inverse columns

but the scenario that we can't do is multiple ranges. i.e. 1-4, 5-6

One idea on how to convert that would be to provide an array where the users would specify the indexes with the convention that they should be treated in groups; example: new [] { 1, 3, 5, 7} would translate to 1-3, 5-7 but that seems UGLY.

Another idea is to let the user specify multiple attributes for the range, and have the column name specified separately. Example:
 [LoadColumn(start:"0", end: "8")]
 [LoadColumn(start:"10", end: "18")]
 [ColumnName( "Features")]
 public float SepalLength;
But, IMHO, the range regex beats this too from the prospective of simplicity and ease of use.

So to summarize, i propose:
1- We keep the old ordinal, as string, rename to range
2- We add a boolean LoadInverseRange to determine whether we add the columns defined in range or they should be the one not loaded (AllOther in TextLoader.Range)

The construction of the TextLoader columns from the attributes is not final. I will complete after getting a final word here.

@eerhardt, @TomFinley, @Zruty0, @CESARDELATORRE

Decision is to break down the range/ordinal. If more granularity is needed, the users can use the text loader columns. #RESOLVED #Resolved

sfilipi · 2018-12-18T07:13:40Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

    {
-        public ColumnAttribute(string ordinal, string name = null)
+
+        public LoadColumnAttribute(int ordinal, string name = null, bool loadAllOthers = false)


public [](start = 8, length = 6)

XMl docs on all ctors and samples for all ctors #Resolved

sfilipi · 2018-12-18T16:07:50Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

+        public string Start { get; }
+        public string End { get; }
+        public int[] ColumnsIndexes { get; }
+        public int[] InverseColumnsIndexes { get; }


not needed, remove #Resolved

Zruty0 · 2018-12-18T19:13:20Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

@@ -68,30 +68,95 @@ public VectorTypeAttribute(params int[] dims)
    /// column encapsulates.
    /// </summary>
    [AttributeUsage(AttributeTargets.Field | AttributeTargets.Property, AllowMultiple = false, Inherited = true)]
-    public sealed class ColumnAttribute : Attribute
+    public sealed class LoadColumnAttribute : Attribute


LoadColumnAttribute [](start = 24, length = 19)

This attribute should go next to TextLoader, not as part of this class. #Closed

Zruty0 · 2018-12-18T19:14:12Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

+        /// <param name="ordinal">The index of the column in the text file.</param>
+        /// <param name="name">The optional name of the column, if it should be different from the name of the field or property where the attribute is positioned.</param>
+        /// <param name="loadAllOthers">Wheather to load all columns, besides the one specified in the ordinal.</param>
+        public LoadColumnAttribute(int ordinal, string name = null, bool loadAllOthers = false)


ordinal [](start = 39, length = 7)

how about index, or columnIndex? The word ordinal does sort-of mean the same thing, but it's the only place in the codebase that we use the word #Closed

Zruty0 · 2018-12-18T19:14:41Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

+        /// <param name="ordinal">The index of the column in the text file.</param>
+        /// <param name="name">The optional name of the column, if it should be different from the name of the field or property where the attribute is positioned.</param>
+        /// <param name="loadAllOthers">Wheather to load all columns, besides the one specified in the ordinal.</param>
+        public LoadColumnAttribute(int ordinal, string name = null, bool loadAllOthers = false)


name [](start = 55, length = 4)

can we remove name? This duplicates the functionality of ColumnName for no good reason #Closed

Zruty0 · 2018-12-18T19:15:14Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

+        /// <param name="end">The ending column index, for the range.</param>
+        /// <param name="name">The optional name of the column, if it should be different from the name of the field or property where the attribute is positioned.</param>
+        /// <param name="columnIndexes">Distinct text file column indces to load as part of this column.</param>
+        public LoadColumnAttribute(string start, string end = null, string name = null, int[] columnIndexes = null)


string [](start = 35, length = 6)

start and end should be integers here. #Resolved

can't have nullable ints. we can't leave 'end' as Auto, if i flip to ints. that ok?

In reply to: 242669241 [](ancestors = 242669241)

If we want to preserve the '10-*' logic, I would go with a different attribute for it ([LoadAllColumns(start: 10)] or something like that).

In reply to: 242741038 [](ancestors = 242741038,242669241)

I don't think the '*' is going to be as useful in API as it is in cmdline.

In reply to: 243120230 [](ancestors = 243120230,242741038,242669241)

remove

In reply to: 243120291 [](ancestors = 243120291,243120230,242741038,242669241)

Zruty0 · 2018-12-18T19:15:49Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

+        /// <param name="end">The ending column index, for the range.</param>
+        /// <param name="name">The optional name of the column, if it should be different from the name of the field or property where the attribute is positioned.</param>
+        /// <param name="columnIndexes">Distinct text file column indces to load as part of this column.</param>
+        public LoadColumnAttribute(string start, string end = null, string name = null, int[] columnIndexes = null)


columnIndexes [](start = 94, length = 13)

I think this parameter is more misleading than it is helpful. Can you remove it? #Closed

Zruty0 · 2018-12-18T19:16:23Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

+        /// </summary>
+        /// <param name="columnIndexes">Distinct text file column indces to load as part of this column.</param>
+        /// <param name="name">The optional name of the column, if it should be different from the name of the field or property where the attribute is positioned.</param>
+        public LoadColumnAttribute(int[] columnIndexes, string name = null)


columnIndexes [](start = 41, length = 13)

columnIndices #Closed

Zruty0 · 2018-12-18T19:18:58Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoaderSaverCatalog.cs

+                var mappingAttr = memberInfo.GetCustomAttribute<LoadColumnAttribute>();
+                var mptr = memberInfo.GetCustomAttributes();
+
+                Contracts.Assert(mappingAttr != null, $"Field or property {memberInfo.Name} is missing the LoadColumn attribute");


LoadColumn [](start = 107, length = 10)

maybe nameof ? #Resolved

Using this comment to ask a meta quesiton: does it make more sense to continue that to throw? Maybe not all the properties of the data model will be decorated with LoadColumn. @Zruty0, @tomfinly

In reply to: 242670646 [](ancestors = 242670646)

Resolving the questions, since the discussion has moved to the other comment.

In reply to: 243040913 [](ancestors = 243040913,242670646)

Zruty0 · 2018-12-18T19:19:12Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoaderSaverCatalog.cs

+                var mappingAttr = memberInfo.GetCustomAttribute<LoadColumnAttribute>();
+                var mptr = memberInfo.GetCustomAttributes();
+
+                Contracts.Assert(mappingAttr != null, $"Field or property {memberInfo.Name} is missing the LoadColumn attribute");


Assert [](start = 26, length = 6)

it should be Check if the user can hit it #Closed

Zruty0 · 2018-12-18T19:21:38Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoaderSaverCatalog.cs

+        /// if one of the row contains "5 2:6 4:3" that's mean there are 5 columns all zero
+        /// except for 3rd and 5th columns which have values 6 and 3</param>
+        /// <param name="trimWhitespace">Remove trailing whitespace from lines</param>
+        public static TextLoader CreateTextReader<TInput>(this DataOperations catalog,


CreateTextReader [](start = 33, length = 16)

This entire code should move out of the catalog and into a static method of TextLoader. The catalogs in general should not contain any meaningful code, they are there to expose the existing public interface in a consistent 'folder-like' structure. #Closed

Zruty0 · 2018-12-18T19:24:28Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoaderSaverCatalog.cs

+                Contracts.Assert(mappingAttr != null, $"Field or property {memberInfo.Name} is missing the LoadColumn attribute");
+
+                var column = new TextLoader.Column();
+                column.Name = mappingAttr.Name ?? memberInfo.Name;


Name [](start = 23, length = 4)

check for ColumnName instead #Closed

Zruty0 · 2018-12-18T19:26:45Z

src/Microsoft.ML.Data/Data/SchemaDefinition.cs

+        /// <param name="ordinal">The index of the column in the text file.</param>
+        /// <param name="name">The optional name of the column, if it should be different from the name of the field or property where the attribute is positioned.</param>
+        /// <param name="loadAllOthers">Wheather to load all columns, besides the one specified in the ordinal.</param>
+        public LoadColumnAttribute(int ordinal, string name = null, bool loadAllOthers = false)


loadAllOthers [](start = 73, length = 13)

that's a misleading name. Maybe invert or complement is more appropriate. But, as I've said offline, I think we can safely get rid of this functionality altogether, it's better than having to explain it :) #Resolved

Do you think this will be useful when the user wants to load all features into one column, and the label in another.

we'd spare them having to look up their cols cardinality, and a concat.

In reply to: 242673339 [](ancestors = 242673339)

still not enough reason, in my opinion

In reply to: 242741558 [](ancestors = 242741558,242673339)

removed

In reply to: 243121936 [](ancestors = 243121936,242741558,242673339)

Zruty0 · 2018-12-18T19:27:46Z

test/Microsoft.ML.Tests/Scenarios/Api/Estimators/DecomposableTrainAndPredict.cs

@@ -29,7 +29,7 @@ void New_DecomposableTrainAndPredict()
            var dataPath = GetDataPath(TestDatasets.irisData.trainFilename);
            var ml = new MLContext();

-            var data = ml.Data.CreateTextReader(TestDatasets.irisData.GetLoaderColumns(), separatorChar: ',')
+            var data = ml.Data.CreateTextReader<IrisData>(separator: ',')


CreateTextReader [](start = 31, length = 16)

maybe ReadFromTextFile here and everywhere in these New_ tests (where appropriate)? #Closed

Zruty0 · 2018-12-18T19:28:54Z

I think we should modify the cookbook samples to reflect this as a suggested way to read text files. #Closed

Zruty0 · 2018-12-18T19:29:14Z

Maybe not all of the examples, but there should be some that use the class definition and not TextLoader.Column array.

In reply to: 448340735 [](ancestors = 448340735)

Zruty0

🕐

…oadColumn attribute.

Zruty0 · 2018-12-19T22:53:13Z

docs/code/MlNetCookBook.md

+	[LoadColumn(3)]
+	public string MaritalStatus { get; set; }
+
+	public string[] AllFeatures { get; set; }


AllFeatures [](start = 17, length = 11)

will this work? I am suspecting that reading this will throw #Pending

it does work because i changed the logic. See my questions about whether we should process only the annotated members/field, to make the data models more usable.

In reply to: 243104636 [](ancestors = 243104636)

Zruty0 · 2018-12-19T22:53:36Z

docs/code/MlNetCookBook.md

+var data = mlContext.Data.ReadFromTextFile<InspectedRow>(dataPath,
+	// First line of the file is a header, not a data row.
+	hasHeader: true
+)		


[](start = 1, length = 1)

please avoid tabs in this file #Resolved

revoking review

Zruty0 · 2018-12-20T00:13:30Z

src/Microsoft.ML.Data/DataLoadSave/Text/LoadColumnAttribute.cs

+        /// <param name="start">The starting column index, for the range.</param>
+        /// <param name="end">The ending column index, for the range.</param>
+        public LoadColumnAttribute(int start, int end)
+             : this(start.ToString()) //REVIEW  this is incorrect, but it is just temporary there, until the Legacy API's TextLoader gets deleted.


//REVIEW this is incorrect, but it is just temporary there, until the Legacy API's TextLoader gets deleted. [](start = 38, length = 108)

// REVIEW:
Also place it somewhere at the beginning of the line #Resolved

Zruty0 · 2018-12-20T00:14:52Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoader.cs

+                var mappingAttr = memberInfo.GetCustomAttribute<LoadColumnAttribute>();
+
+                if(mappingAttr == null)
+                    continue;


hmm, no longer throwing? I'm not sure I'm a fan of it. #Resolved

@TomFinley @glebuk what do you think?

does it make more sense to continue that to throw? Maybe not all the properties of the data model will be decorated with LoadColumn. If we throw, we restrict usage by expecting the class to be dedicated to just loading the data, whereas if we don't the type can be used for other thing, and the decorated fields can be used for loading.

In reply to: 243120594 [](ancestors = 243120594)

I agree with you Senja, is there some benefit with throwing?

In reply to: 243358643 [](ancestors = 243358643,243120594)

If we throw, we restrict usage by expecting the class to be dedicated to just loading the data, whereas if we don't the type can be used for other thing, and the decorated fields can be used for loading.

Interesting @sfilipi. So, it strikes me that there are two possible dangers we must guard against.

A user attempts to use a class to define a schema, without proper attribution by mistake. We know this happens a lot.

A user attempts to use a class to define a schema, with proper attribution, but wants to have some additional "extra" properties free for no particular reason.

Like I said I know 1 comes up, and throwing has helped people to guard against misuse. 2 has not been possible for a while now -- in fact, AFAIK was never possible. In fact, if we consider all prior usage of the API (where we follow this pattern), and all prior usage of the text loader, and I think of all the questions that we've gotten about them over the years, both outside of Microsoft in ML.NET and inside Microsoft with our internal tool, I don't recall that anyone has ever run into a situation where this presented a serious barrier to them. (But of course I am not omniscient, and while I read I think most user feedback, I have not read all of it. Have we received requests for this? I am finding it difficult to imagine a concrete scenario.)

Even if the scenario were somehow to come up, I would prefer to solve it by introducing another attribute to indicate that this property is not to be considered by the loader, rather than to have this behavior. So, something like, [NoLoader] or whatnot. (I don't insist on the name. I also don't want us to add it now, until there's a concrete request for it, since my suspicion is that it will never come up.)

For this reason, I disagree with @sfilipi and @singlis and agree with @Zruty0. It must throw. #Resolved

Zruty0 · 2018-12-20T00:21:36Z

test/Microsoft.ML.Tests/Scenarios/Api/Estimators/TrainWithInitialPredictor.cs

@@ -22,7 +22,7 @@ public void New_TrainWithInitialPredictor()

            var ml = new MLContext(seed: 1, conc: 1);

-            var data = ml.Data.CreateTextReader(TestDatasets.Sentiment.GetLoaderColumns(), hasHeader: true).Read(GetDataPath(TestDatasets.Sentiment.trainFilename));


GetLoaderColumns [](start = 71, length = 16)

do we still use GetLoaderColumns() anywhere? Maybe we can delete the method? #Resolved

we do use it in a few tests.

In reply to: 243121856 [](ancestors = 243121856)

Zruty0

positioning comments in the LoadColumnAttribute.

singlis · 2018-12-20T20:33:16Z

src/Microsoft.ML.Data/DataLoadSave/Text/LoadColumnAttribute.cs

+namespace Microsoft.ML.Data
+{
+#pragma warning disable 618
+    /// <summary>


Why is the warning disable needed? I would document why its disabled. #Resolved

singlis · 2018-12-20T20:42:55Z

src/Microsoft.ML.Data/DataLoadSave/Text/LoadColumnAttribute.cs

+        /// Initializes new instance of <see cref="LoadColumnAttribute"/>.
+        /// </summary>
+        /// <param name="columnIndexes">Distinct text file column indices to load as part of this column.</param>
+        // REVIEW: Calling the private constructor with just the columnIndexes[0] parameter, is incorrect,


Can all the "REVIEW" items be tagged with a github issue? Just thinking that maybe easier to track when the Legacy API's TextLoader is deleted. #Resolved

singlis · 2018-12-20T20:43:57Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoader.cs


            [Argument(ArgumentType.AtMostOnce,
                HelpText = "Number of source columns in the text data. Default is that sparse rows contain their size information.",
                ShortName = "size")]
            public int? InputSize;

            [Argument(ArgumentType.AtMostOnce, Visibility = ArgumentAttribute.VisibilityType.CmdLineOnly, HelpText = "Source column separator. Options: tab, space, comma, single character", ShortName = "sep")]
-            public string Separator = "tab";
+            public string Separator = "tab"; //DefaultArguments.Separator


DefaultArguments [](start = 47, length = 16)

Is this comment needed? #Resolved

or maybe just two default arguments Separator and SeparatorChars? Two different params doesnt guarantee things to be in sync. Im assuming since these are argruments that they can be set by the user -- so what happens if a user specifies Separator to be space, but doesnt set the separator chars?

In reply to: 243411408 [](ancestors = 243411408)

singlis · 2018-12-20T21:17:43Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoaderSaverCatalog.cs

-            Column[] columns, bool hasHeader = false, char separatorChar = '\t', IMultiStreamSource dataSample = null)
+            TextLoader.Column[] columns,
+            bool hasHeader = false,
+            char separatorChar = '\t',


separatorChar = TextLoader.DefaultArguments.Separator? #Resolved

singlis · 2018-12-20T21:25:17Z

src/Microsoft.ML.Data/DataLoadSave/Text/TextLoaderSaverCatalog.cs

+        public static void SaveAsText(this DataOperations catalog,
+            IDataView data,
+            Stream stream,
+            char separator = '\t',


DefaultArguments.Separator. #Resolved

singlis

sfilipi added 2 commits December 11, 2018 15:59

moving reflection-based read out of the legacy project.

f699cb0

adding more members to LoadColumnAttribute

3931d13

sfilipi requested review from Ivanidzo4ka, CESARDELATORRE, Zruty0, TomFinley and eerhardt December 14, 2018 00:23

sfilipi self-assigned this Dec 14, 2018

sfilipi added the API Issues pertaining the friendly API label Dec 14, 2018

sfilipi added this to the 1218 milestone Dec 14, 2018

Ivanidzo4ka changed the title ~~[WIP] Schema based text laoder~~ [WIP] Schema based text loader Dec 14, 2018

Ivanidzo4ka reviewed Dec 14, 2018

View reviewed changes

wschin reviewed Dec 14, 2018

View reviewed changes

sfilipi added 2 commits December 17, 2018 23:03

Breaking down the ordinal/range into ordinal, start, end, columnIndices.

7cacf58

Introducing test.

reverting updating the libmf module.

c69eb4b

sfilipi changed the title ~~[WIP] Schema based text loader~~ Schema based text loader Dec 18, 2018

sfilipi commented Dec 18, 2018

View reviewed changes

Removing the option to have ranges + allOthers together.

31e17d9

Zruty0 reviewed Dec 18, 2018

View reviewed changes

Zruty0 previously requested changes Dec 18, 2018

View reviewed changes

sfilipi added 4 commits December 19, 2018 12:57

Addressing most comments

e1201bb

Merge branch 'master' into schemaBasedReader

e583b88

post merge fixes

55e1bdd

this test is not needed if we don't process data models without the L…

01fab7a

…oadColumn attribute.

Zruty0 reviewed Dec 19, 2018

View reviewed changes

Zruty0 reviewed Dec 20, 2018

View reviewed changes

Zruty0 approved these changes Dec 20, 2018

View reviewed changes

removing tabs from the cookbook.

c540d7f

positioning comments in the LoadColumnAttribute.

singlis reviewed Dec 20, 2018

View reviewed changes

singlis approved these changes Dec 20, 2018

View reviewed changes

sfilipi added 2 commits December 21, 2018 09:04

addressing feedback

ce25b69

addign back the test about the exception. Reverting the changes to libmf

0392712

sfilipi merged commit c4236e3 into dotnet:master Dec 21, 2018

sfilipi deleted the schemaBasedReader branch December 21, 2018 17:58

eerhardt mentioned this pull request Sep 27, 2019

[DatabaseLoader] Error when using attributes (i.e ColumnName) #4195

Closed

ghost locked as resolved and limited conversation to collaborators Mar 26, 2022

		@@ -22,7 +22,7 @@ public void New_TrainWithInitialPredictor()

		var ml = new MLContext(seed: 1, conc: 1);

		var data = ml.Data.CreateTextReader(TestDatasets.Sentiment.GetLoaderColumns(), hasHeader: true).Read(GetDataPath(TestDatasets.Sentiment.trainFilename));

Schema based text loader #1878

Schema based text loader #1878

Conversation

sfilipi commented Dec 14, 2018 • edited Loading

Ivanidzo4ka Dec 14, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wschin Dec 14, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

sfilipi commented Dec 18, 2018 • edited Loading

sfilipi commented Dec 18, 2018 • edited Loading

sfilipi Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

sfilipi Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited by sfilipi Loading

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

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 Dec 18, 2018 • edited by sfilipi Loading

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

Zruty0 Dec 18, 2018 • edited Loading

Choose a reason for hiding this comment

Zruty0 commented Dec 18, 2018 • edited Loading

Zruty0 commented Dec 18, 2018

Zruty0 left a comment

Choose a reason for hiding this comment

Zruty0 Dec 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Zruty0 Dec 19, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Zruty0 Dec 20, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Zruty0 Dec 20, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

sfilipi Dec 20, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TomFinley Dec 21, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Zruty0 Dec 20, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Zruty0 left a comment

Choose a reason for hiding this comment

singlis Dec 20, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

singlis Dec 20, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

singlis Dec 20, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis Dec 20, 2018 • edited by sfilipi Loading

Choose a reason for hiding this comment

sfilipi commented Dec 14, 2018 •

edited

Loading

Ivanidzo4ka Dec 14, 2018 •

edited by sfilipi

Loading

wschin Dec 14, 2018 •

edited by sfilipi

Loading

sfilipi commented Dec 18, 2018 •

edited

Loading

sfilipi commented Dec 18, 2018 •

edited

Loading

sfilipi Dec 18, 2018 •

edited

Loading

sfilipi Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited by sfilipi

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited by sfilipi

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 Dec 18, 2018 •

edited by sfilipi

Loading

Zruty0 Dec 18, 2018 •

edited

Loading

Zruty0 commented Dec 18, 2018 •

edited

Loading

Zruty0 Dec 19, 2018 •

edited by sfilipi

Loading

Zruty0 Dec 19, 2018 •

edited by sfilipi

Loading

Zruty0 Dec 20, 2018 •

edited by sfilipi

Loading

Zruty0 Dec 20, 2018 •

edited by sfilipi

Loading

sfilipi Dec 20, 2018 •

edited

Loading

TomFinley Dec 21, 2018 •

edited by sfilipi

Loading

Zruty0 Dec 20, 2018 •

edited by sfilipi

Loading

singlis Dec 20, 2018 •

edited by sfilipi

Loading

singlis Dec 20, 2018 •

edited by sfilipi

Loading

singlis Dec 20, 2018 •

edited by sfilipi

Loading

singlis Dec 20, 2018 •

edited by sfilipi

Loading

singlis Dec 20, 2018 •

edited by sfilipi

Loading