bhrnjica / anndotnet Goto Github PK

Confusion matrix dialog window has wrong title. It stand "Confusion matrix for Training data set", event if the confusion matrix is opened for validation data set.

This modul will provide complete GUI solution from image preparation to testing model. This include implementation of convolution and polling layers in the Network Designer.

Evaluate button in Test tab page generates a new result row for every button press, even for the same row.

Project explorer is showing loaded projects and ML configurations in tree structure. User should have possibility to righ click on any tree item and context menu with options should be shown.
Currently the flowing options are supported on mlconfig tree item:

• Rename,
• Delete.

More options should be added:

• duplicate- this menu option should duplicate selected mlconfig create a new one with the same configuration. Once it is duplicated it should be added to the annproject. This option is useful when the user want to make a copy mlconfig in order to change something minor.
• Rename, - for project tree item.
  -more options

When I do the following:

1. Clone solution by typing: git clone https://github.com/bhrnjica/anndotnet
2. Run Visual Studio 2017 and open anndotnet.sln solution file.
3. Set anndotnet.tool – as the startup project

The anndotnet.tool.net is shown "unavailable" and it is not possible to Set anndotnet.tool – as the startup project.
The solution cannot be Rebuilt.
How do I get started with anndotnet.tool?
How do I load the sln into VS 2017 and rebuild it?

• ANNdotNET vNext should be able to load huge dataset (more than milion rows) in ANN Project.
• MLConfig creation from huge dataset should also be optimized.
• Provide progress report for huge MLConfig creation

ANNdotNET Feed is component places at the StartPage and contains list of examples which is dynamically populated every time user start the application. The Feed has central place at github.com/bhrnjica/anndotnet/Examples folder, it contains all examples allowed to appear in the Feed.

With this way the Feed is not dependent of the version, and dynamically shows examples in various ML fields.

The user will have ability to see description of each example and by double click download the whole example and load into the ANNdotNET just like other examples at the Start Page.

This component will try to extend collaboration between users and each user will have possibility to share his interesting projects.

The feature looks like:

On this way unnecessary files are collecting in the model folder, without reason, and they should be deleted.
Once the training session is over, the best model should be created in the model folder, and no other model files should be there.

When creating an new mlcofnig file, the exception is thrown. And the model cannot be trained.

Provide possibility of creation type of layer to support more than one LSTM layer in the network.

Currently only

• Start Page
  -Binary Model Performance
  -Multi-class Model Performance

dialogs are based on WinForm.

The annproject contains two Tab pages.

• Data - contains rawdataset
• Info - should contains relevant information about the project. Citation for the database and related copyrighted work.

The chart control text is very small, and should be re-scaled. Also there are some information which should be also removed, in order to simplify the controls.

To quick fix, define validation data set with at least one row:

Seems the wrong DocumentFormat.OpenXml.dll version is deployed on the latest version. During the export option, the Exception is throws with the message taht DocumentFormat.OpenXml.dll 2.7.2 should be presented.

Upload the project on GitHub, and try to stabilize the code by implementing and running tests, and bug fixing through the reported issues.

Using Graphviz CNTK model can be visualized in Python, but the feature is not available in C#.

Upgrade MLEngine from CNTK 2.5.1 to CNTK 2.6, after the mayor issue is identified and solved.

In the data section I set the validation data set to 10%, train the model go to Evaluation tab and refresh but only get Training Vs Predicted output, the Validation set seems to be empty.
Is this a bug or am I doing something stupid!

Prepare and transform code to support .NET Core 3

After the mlconfiguration tree item is renamed, name changes are not updated in to annproject file.

"I downloaded the executable version of ANNdotNET 1.2. I'm pretty sure I did the image preparation correctly with the transformation information 32,32,3 and Use Augmentation. Saved that. Next, in network settings, I defined the VND as shown. Saved that. But then I noticed the Input Layer: Image Shape was showing [0,32,32]. Going back to Image Preparation, it showed 32,32,0. Changing the 0 to 3 produced a red (error) box and did not change the network settings. Finally I closed out of the App and literally edited the CIFAR_rawdata.txt and the CIFAR-10-ConvNet.mlconfig files (changing the 0 dim to 3). Amazingly that worked.
The ANNdotNET project file loaded and image shape shows [3,32,32]. Training appears to be running fine."
T

Implement set of tests and example to test new features in ANNdotNET 1.2.

When create new Default Project the following binding errors accrued in the debug window:

System.Windows.Data Error: 40 : BindingExpression path error: 'UseActivation' property not found on 'object' ''NNLayer' (HashCode=24723024)'. BindingExpression:Path=UseActivation; DataItem='NNLayer' (HashCode=24723024); target element is 'Label' (Name='lblAct'); target property is 'IsEnabled' (type 'Boolean')
System.Windows.Data Error: 40 : BindingExpression path error: 'Network' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=Network; DataItem='ANNProjectController' (HashCode=18965813); target element is 'ListBox' (Name='imageLabelList'); target property is 'ItemsSource' (type 'IEnumerable')
System.Windows.Data Error: 40 : BindingExpression path error: 'SelectedIndex' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=SelectedIndex; DataItem='ANNProjectController' (HashCode=18965813); target element is 'ListBox' (Name='imageLabelList'); target property is 'SelectedIndex' (type 'Int32')
System.Windows.Data Error: 40 : BindingExpression path error: 'Width' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=Width; DataItem='ANNProjectController' (HashCode=18965813); target element is 'TextBox' (Name='txtHdim2'); target property is 'Text' (type 'String')
System.Windows.Data Error: 40 : BindingExpression path error: 'Height' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=Height; DataItem='ANNProjectController' (HashCode=18965813); target element is 'TextBox' (Name='txtHdim12'); target property is 'Text' (type 'String')
System.Windows.Data Warning: 40 : BindingExpression path error: 'Channels' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=Channels; DataItem='ANNProjectController' (HashCode=18965813); target element is 'TextBox' (Name='txtHdim13'); target property is 'Text' (type 'String')
System.Windows.Data Error: 40 : BindingExpression path error: 'IsAugmentation' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=IsAugmentation; DataItem='ANNProjectController' (HashCode=18965813); target element is 'CheckBox' (Name=''); target property is 'IsChecked' (type 'Nullable`1')

The bidnign errors are related for ImageClassifier Windows, since this is empty when default prject is loaded. This binding errors should be removed.

If the used switch from project to mlconfig the following binding errors occurred:

System.Windows.Data Error: 40 : BindingExpression path error: 'UseActivation' property not found on 'object' ''NNLayer' (HashCode=885082)'. BindingExpression:Path=UseActivation; DataItem='NNLayer' (HashCode=885082); target element is 'Label' (Name='lblAct'); target property is 'IsEnabled' (type 'Boolean')

System.Windows.Data Error: 40 : BindingExpression path error: 'Network' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=Network; DataItem='ANNProjectController' (HashCode=18965813); target element is 'ListBox' (Name='imageLabelList'); target property is 'ItemsSource' (type 'IEnumerable')
System.Windows.Data Error: 40 : BindingExpression path error: 'SelectedIndex' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=SelectedIndex; DataItem='ANNProjectController' (HashCode=18965813); target element is 'ListBox' (Name='imageLabelList'); target property is 'SelectedIndex' (type 'Int32')
System.Windows.Data Error: 40 : BindingExpression path error: 'Width' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=Width; DataItem='ANNProjectController' (HashCode=18965813); target element is 'TextBox' (Name='txtHdim2'); target property is 'Text' (type 'String')
System.Windows.Data Error: 40 : BindingExpression path error: 'Height' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=Height; DataItem='ANNProjectController' (HashCode=18965813); target element is 'TextBox' (Name='txtHdim12'); target property is 'Text' (type 'String')
System.Windows.Data Warning: 40 : BindingExpression path error: 'Channels' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=Channels; DataItem='ANNProjectController' (HashCode=18965813); target element is 'TextBox' (Name='txtHdim13'); target property is 'Text' (type 'String')
System.Windows.Data Error: 40 : BindingExpression path error: 'IsAugmentation' property not found on 'object' ''ANNProjectController' (HashCode=18965813)'. BindingExpression:Path=IsAugmentation; DataItem='ANNProjectController' (HashCode=18965813); target element is 'CheckBox' (Name=''); target property is 'IsChecked' (type 'Nullable`1')
System.Windows.Data Error: 40 : BindingExpression path error: 'UseActivation' property not found on 'object' ''NNLayer' (HashCode=65623433)'. BindingExpression:Path=UseActivation; DataItem='NNLayer' (HashCode=65623433); target element is 'Label' (Name='lblAct'); target property is 'IsEnabled' (type 'Boolean')

The last binding errors are related for Visual Network Designer

Once the model is trained, sometimes the right model is not evaluated. Also Export option cannot find the right path of the model.

Add more ann layers in the NNetwork Designer.

• CNTK Cuda optimized layers for RNN (Stacked LSTM and GRU)
• NALU-Neural Arithmetic Logic Units layer

In case the folder of mlconfig file is used by another process, renaming mlconfig name cause whole annproject to be corrupted.

Version v.10 supports for splitting raw data set into two data sets : Train and Validation. The next version should support for three data sets.
Validation and Test data sets are optional and the user is not forced to created them them in case he doesn't want them.

Write blog post about Image Classification in order to provide complete example how create and train Image Classification deep learning model in ANNdotNET.

Write blog post about ANNdotNET Feed Example, and other features.

Currently only Training and Validation sets are exported to excel. Since the next version will support three datasets, export to excel should also provide those datasets.

This is caused by new feature of automatic evaluation of the model, when the user switch to evaluate tab page.
For now please don't try to switch to evaluate tab page until the bug fix released. Hope this will be soon.

Seem that not all WPF components and included external resources (projectinfo.rtf file) are not disposed 100%. Some memory still remain after project is closed. I think the main problem is projectinfo.rtf file loaded into rich-text control. So disposing richtext content from rtf content should fix the problem in most but not all cases.

1. Randomize raw dataset and then split into training and validation data set
2. Split first raw data set on training and validation and then randomizes training data set only. This can be achieved enabling randomization of mini-batch during training.

Officially ANNdotNET can run on Windows 10, since ANNdotNET strongly relies on CNTK. However I have received many mails, and questions is it possible to run it on Win8 and Wind7.

Recently I have performed some tests to check if ANNdotNET can run on Windows 8.1 and Windows 7. The test result are positive.

I could successfully run ANNdotNET v1.1 on Windows 8.1 and Window 7 64 bit OS machines.

The tests were performed on clean Windows 8.1 and Windows 7 machines.
Prior to run, I had to install the following software requirements, which are mentioned in the Software Requirements:

• .NET Framework 4.7.2. (http://go.microsoft.com/fwlink/?linkid=863262)
• Visual C++ 2017 version 15.4 v14.11 toolset, (https://aka.ms/vs/15/release/vc_redist.x64.exe)
• Visual C++ Redistributable Packages for Visual Studio 2013. Only vcredist_x64.exe file should be downloaded and installed. (https://www.microsoft.com/en-us/download/details.aspx?id=40784)

Feel free to add any comment or issue in case you have different experience during installation.

• Rename,
• Duplicate and
• Delete
  context menu options

When continue training option is checked, history training graphs are clear and started without previous training history data.

After closing project, mlconfig file is saved the Result parameter with invalid file name.

Currently timeseries rawData must contains only one column numbers witouth header and other columns. Implement support for header and more than one column

Famous Kaggle Image Classification problem Cat&Dog will be added as full example to ANNdotNET, and to be example how to make image classifications problem in ANNdotNET.

This is standard way of making long running process more interesting to the user while waiting for the results.

Find below the crash screenshot after loading SentimentAnalysys.zip example. It appears clicking RUN.
I'm using ANNdotNET v1.2-rc20191602

When running ANNdotNET on PCs with different localization settings (precisely with European localization) there are some issues when exporting and displaying results.