| 1925fb8 | moneta | 20 November 2017, 10:23:58 UTC | Add a fast implementation of Im2Col computing first the matrix mapping in a vector and then use the vector to copy from input to output matrix - add option to use MT with the TThreadExecutor, but found using 4 threads is 10% slower and with 4+4 threads is 20% slower. use then by default no MT | 20 November 2017, 10:23:58 UTC |
| 290f8d3 | moneta | 16 November 2017, 17:20:35 UTC | some Speed up in im2col and convloutional operation Do not use TThreadExecutor | 16 November 2017, 17:20:35 UTC |
| 3884209 | moneta | 16 November 2017, 15:44:40 UTC | Use TThreadExecutor::Foreach since we do not need to return a vector in the map operation | 16 November 2017, 15:44:40 UTC |
| 906d8f8 | moneta | 16 November 2017, 15:06:04 UTC | Merge changes happened in the master before creating tmva-dnn branch - change in TMVA_input = add TThreadExecutor in TMVA::Config | 16 November 2017, 15:06:04 UTC |
| 41900ab | moneta | 16 November 2017, 10:31:42 UTC | Fix for weight file name when we do not save the model | 16 November 2017, 10:50:06 UTC |
| 5074198 | moneta | 14 November 2017, 16:16:50 UTC | Fix a bug in doing the minimisation step. use now all the data and not only one batch. Also add the capability to save the model weights every time a better test error is found | 16 November 2017, 10:50:06 UTC |
| 01e6124 | moneta | 14 November 2017, 13:48:27 UTC | Add more debug and measure time also for computing training and test error from Loss function | 16 November 2017, 10:50:06 UTC |
| 43ebf0a | moneta | 14 November 2017, 13:47:20 UTC | Get number of CPU in TMVA::Config | 16 November 2017, 10:50:06 UTC |
| dcc824d | moneta | 14 November 2017, 10:52:56 UTC | Speed-up evaluation of TTreeFormula in DataSetFactory (use TFormula quick mode) | 16 November 2017, 10:50:06 UTC |
| 41f29e7 | moneta | 14 November 2017, 09:14:42 UTC | Add new option MaxEpochs to stop minimisation after the maximum number of epochs is reached | 16 November 2017, 10:50:06 UTC |
| cf9f7e5 | moneta | 13 November 2017, 18:18:38 UTC | Remove computation of correlation matrix and remove check for constant variables | 16 November 2017, 10:50:06 UTC |
| a5ef244 | moneta | 10 November 2017, 08:12:06 UTC | Fix the padding when computing the activation gradient | 16 November 2017, 10:50:05 UTC |
| 7e8b5c4 | moneta | 09 November 2017, 17:26:53 UTC | Remove some debug print out and improve test | 16 November 2017, 10:50:05 UTC |
| a25bda1 | moneta | 09 November 2017, 17:02:41 UTC | Thanks to Vladimir fix weight gradient and activation gradient computation in conv. layer. Fix also backward pass in the maxpool layer Add Assert in matrix operations Add a test for backward pass comparing weight gradient with those computed with finite differences | 16 November 2017, 10:50:05 UTC |
| 5b10db8 | moneta | 25 October 2017, 12:36:35 UTC | Fix input parameter for Reshape Layer | 16 November 2017, 10:50:05 UTC |
| eefe1b4 | moneta | 25 October 2017, 07:33:37 UTC | Fix layout string for DNN test. Need a reshape layer before a DNN layer | 16 November 2017, 10:50:05 UTC |
| 1487b80 | moneta | 24 October 2017, 08:31:13 UTC | Fix test file name and input batch layout | 16 November 2017, 10:50:05 UTC |
| 812ad9a | Omar Zapata | 24 October 2017, 09:33:25 UTC | TMVA: removed more warnings | 16 November 2017, 10:50:05 UTC |
| 0f540a8 | Omar Zapata | 24 October 2017, 09:05:47 UTC | TMVA: removing more warnings from multiple types of layers and in some tests | 16 November 2017, 10:50:05 UTC |
| 2d54c48 | Omar Zapata | 24 October 2017, 07:46:08 UTC | TMVA: removed warnings in TensorDataLoader and TestBackpropagationDL | 16 November 2017, 10:50:05 UTC |
| 7e8d670 | Omar Zapata | 24 October 2017, 07:29:42 UTC | TMVA: remove warnings in DenoisePropagation.cxx and Propagation.cxx | 16 November 2017, 10:50:05 UTC |
| c90fbf9 | Saurav Shekhar | 20 October 2017, 12:13:36 UTC | FIX: MVaValue Calculation in Cpu Architecture | 16 November 2017, 10:50:05 UTC |
| 3ae5d89 | Omar Zapata | 13 October 2017, 07:57:58 UTC | TMVA: moved fPool from TCpuMatrix to TMVA::Config class and removed more warnings | 16 November 2017, 10:50:05 UTC |
| c927cdf | Omar Zapata | 12 October 2017, 20:29:30 UTC | TMVA: removed compilation warnings | 16 November 2017, 10:50:04 UTC |
| 73bc91a | Omar Zapata | 12 October 2017, 15:29:33 UTC | TMVA: * removed debug messages * fixed dummy layer in TDeepNet::Backward that producing segfault. * added initialization to zero in TCpuMatrix * set batch size to 1 in GetMvaValue in MethodDL | 16 November 2017, 10:50:04 UTC |
| 283e338 | Omar Zapata | 12 October 2017, 13:52:24 UTC | TMVA: implemented method GetMvaValue for method DL | 16 November 2017, 10:50:04 UTC |
| 169210e | Saurav Shekhar | 05 October 2017, 13:19:18 UTC | Minor change params of RNNLayer | 16 November 2017, 10:50:04 UTC |
| 3001f79 | Saurav Shekhar | 05 October 2017, 10:45:26 UTC | Minor changes, methodDL multi-threading in Minimizer removed | 16 November 2017, 10:50:04 UTC |
| 7ba209e | Saurav Shekhar | 04 October 2017, 08:46:59 UTC | Added Cuda Support in recurrent propagation | 16 November 2017, 10:50:04 UTC |
| 7f969c8 | Saurav Shekhar | 29 August 2017, 04:13:27 UTC | CPU (Blas) Support added | 16 November 2017, 10:50:04 UTC |
| 7d93a8e | Saurav Shekhar | 29 August 2017, 01:48:40 UTC | RNN dimensions changed and full network working | 16 November 2017, 10:50:04 UTC |
| a838981 | Saurav Shekhar | 25 August 2017, 10:07:30 UTC | MethodDL RNN Parser added | 16 November 2017, 10:50:04 UTC |
| 5b0d943 | IlievskiV | 23 August 2017, 10:17:07 UTC | fix: Initialize bias gradients to zero | 16 November 2017, 10:50:04 UTC |
| a9ee00e | IlievskiV | 22 August 2017, 22:15:28 UTC | test: Add test for Method DL, for DNN case | 16 November 2017, 10:50:04 UTC |
| 2754629 | IlievskiV | 22 August 2017, 15:02:10 UTC | test: Add test for Method DL, for the DNN case | 16 November 2017, 10:50:04 UTC |
| 1f2e051 | IlievskiV | 22 August 2017, 14:38:34 UTC | feat: Add additional condition for loading batches When the first layer is Dense, the batch is only one matrix, where each row is one example. | 16 November 2017, 10:50:04 UTC |
| 65178fa | IlievskiV | 22 August 2017, 13:59:30 UTC | feat: Define batch layout string Add another string for the batch dimensions to be specified at the user input. | 16 November 2017, 10:50:04 UTC |
| 19d3e96 | IlievskiV | 22 August 2017, 13:28:35 UTC | fix: Add condition for dummy backward gradients in the Dense Layer | 16 November 2017, 10:50:03 UTC |
| b3451b9 | Saurav Shekhar | 17 August 2017, 13:36:37 UTC | feat: Backprop test for Denselayer added | 16 November 2017, 10:50:03 UTC |
| ac0b5ce | IlievskiV | 15 August 2017, 19:07:05 UTC | fix: Multiply Transponse errot for CPU backend | 16 November 2017, 10:50:03 UTC |
| 0744cb2 | IlievskiV | 15 August 2017, 19:05:54 UTC | test: Add test for testing Method DL for CPU | 16 November 2017, 10:50:03 UTC |
| bf9bf0b | IlievskiV | 14 August 2017, 07:40:58 UTC | feat: Define input layout string Definition of the input layout string and the appropriate parsing. This string is defining the deimensions of the input. | 16 November 2017, 10:50:03 UTC |
| 0f6a36b | IlievskiV | 13 August 2017, 17:09:39 UTC | test: Add tests for minimizers Include tests for Step, Step Momentum and Step Nesterov methods | 16 November 2017, 10:50:03 UTC |
| fe2041c | IlievskiV | 13 August 2017, 17:08:50 UTC | feat: Add test for loading real dataset | 16 November 2017, 10:50:03 UTC |
| e2bb747 | IlievskiV | 13 August 2017, 17:05:12 UTC | fix: Change to reference input in the Forward call | 16 November 2017, 10:50:03 UTC |
| 166d0e7 | IlievskiV | 13 August 2017, 16:49:23 UTC | fix: Fix Conv Layer Backward | 16 November 2017, 10:50:03 UTC |
| 83daf18 | Saurav Shekhar | 09 August 2017, 12:41:06 UTC | temp: Full RNN fixes | 16 November 2017, 10:50:03 UTC |
| 49c0a46 | IlievskiV | 09 August 2017, 11:04:55 UTC | fix: Bug fix in the Conv Layer Backprop step | 16 November 2017, 10:50:03 UTC |
| 354ef39 | IlievskiV | 09 August 2017, 09:20:52 UTC | feat: Add flattening option in the Reshape Layer | 16 November 2017, 10:50:03 UTC |
| bef18a6 | IlievskiV | 08 August 2017, 14:17:26 UTC | test: Add test for Flatten for the Reference backend | 16 November 2017, 10:50:03 UTC |
| 36467e0 | IlievskiV | 08 August 2017, 13:33:27 UTC | test: Add Tensor Data Loader test for CPU backend | 16 November 2017, 10:50:02 UTC |
| 033fea1 | IlievskiV | 08 August 2017, 13:31:40 UTC | feat: Implement Flatten and Delfatten for Reference and CPU | 16 November 2017, 10:50:02 UTC |
| db331e4 | IlievskiV | 08 August 2017, 13:30:42 UTC | feat: Define Flatten and Deflatten kernels | 16 November 2017, 10:50:02 UTC |
| bbb0dc9 | IlievskiV | 08 August 2017, 08:52:16 UTC | test: Add test for Tensor Data Loader for Reference backend Implemented test for the Tensor Data Loader for the Reference Backend with a Tensor Input as a type of input. | 16 November 2017, 10:50:02 UTC |
| 0a02d0d | IlievskiV | 08 August 2017, 08:51:52 UTC | feat: Add function for constructing linear conv net | 16 November 2017, 10:50:02 UTC |
| 32c66ee | IlievskiV | 08 August 2017, 08:50:49 UTC | fix: Input Tensor not initialized properly | 16 November 2017, 10:50:02 UTC |
| e1e765a | IlievskiV | 07 August 2017, 13:42:25 UTC | feat: Implement Tensor Data Loader for Reference Provided a specialized implementation for the Tensor Data Loader for the Reference Architecture. | 16 November 2017, 10:50:02 UTC |
| f426b55 | IlievskiV | 07 August 2017, 10:22:20 UTC | test: Add Conv Net Prediction function test for CPU | 16 November 2017, 10:50:02 UTC |
| c6faf39 | IlievskiV | 07 August 2017, 10:20:39 UTC | test: Add Conv Net Loss function test for CPU | 16 November 2017, 10:50:02 UTC |
| cca0440 | IlievskiV | 07 August 2017, 10:04:12 UTC | test: Add Conv Forward Pass Test for CPU | 16 November 2017, 10:50:02 UTC |
| cc0a162 | IlievskiV | 07 August 2017, 08:09:27 UTC | test: Add Im2Col, Downsample and RotateWeights tests for CPU | 16 November 2017, 10:50:02 UTC |
| 474be40 | ajatgd | 06 August 2017, 18:37:28 UTC | Removing Layer Type attribute from general layer and adding docs for some autoencoder layers | 16 November 2017, 10:50:02 UTC |
| 7ec9cec | Saurav Shekhar | 06 August 2017, 15:27:51 UTC | Full example added | 16 November 2017, 10:50:02 UTC |
| aef024d | IlievskiV | 05 August 2017, 17:05:35 UTC | feat: Implement Copy function in Tensor Data Loader | 16 November 2017, 10:50:02 UTC |
| 67542c3 | IlievskiV | 05 August 2017, 17:04:54 UTC | feat: Implement the CPU architecture for Conv Layers | 16 November 2017, 10:50:01 UTC |
| b5f1622 | IlievskiV | 05 August 2017, 17:03:35 UTC | refactor: Format the coding style | 16 November 2017, 10:50:01 UTC |
| 4fcae31 | ajatgd | 03 August 2017, 10:08:04 UTC | Adding an attribute for the type of layer in General Layer | 16 November 2017, 10:50:01 UTC |
| dd034fb | ajatgd | 02 August 2017, 14:08:00 UTC | Adding FineTune function in DeepNet and test for same | 16 November 2017, 10:50:01 UTC |
| a0d9a87 | Saurav Shekhar | 01 August 2017, 16:00:57 UTC | Forward test RNN added | 16 November 2017, 10:50:01 UTC |
| ffd9222 | ajatgd | 01 August 2017, 11:53:58 UTC | Refactor: Adding two parameters to Backward in all layers | 16 November 2017, 10:50:01 UTC |
| 0946ccc | ajatgd | 31 July 2017, 17:13:19 UTC | Refactor: Migrating layers to new general layer constructor, adding deepnet test for autoencoder | 16 November 2017, 10:50:01 UTC |
| fa61e87 | ajatgd | 28 July 2017, 02:08:40 UTC | Refactor: Adding modified Layers to DeepNet and adding pretrain | 16 November 2017, 10:50:01 UTC |
| b915305 | ajatgd | 28 July 2017, 02:02:10 UTC | Refactor: Adding Corruption, Compression, Reconstruction layer in accordance with General Layer and removing Denoise Layer | 16 November 2017, 10:50:01 UTC |
| c000b14 | Saurav Shekhar | 27 July 2017, 23:49:34 UTC | refactor: pointers removed from ScaleAdd and Copy signatures | 16 November 2017, 10:50:01 UTC |
| 03f164a | sshekh | 27 July 2017, 07:31:45 UTC | refactor: Migrate to vector of weights and biases, DAE Build Breaking In General Layer we have to keep vector of weights and matrices. | 16 November 2017, 10:50:01 UTC |
| 88e3db3 | ajatgd | 25 July 2017, 09:58:27 UTC | Adding Logistic Regression Layer to DeepNet | 16 November 2017, 10:50:01 UTC |
| 4c04ae4 | ajatgd | 25 July 2017, 09:57:14 UTC | Adding tests for Logistic Regression Layer | 16 November 2017, 10:50:01 UTC |
| 44a82a3 | ajatgd | 25 July 2017, 09:54:30 UTC | Adding Logistic Regression Layer and removing Transformed Layer as it is added in Denoise Layer | 16 November 2017, 10:50:00 UTC |
| 99ba388 | ajatgd | 23 July 2017, 21:07:41 UTC | Adding Denoise Layers to DeepNet | 16 November 2017, 10:50:00 UTC |
| b113144 | ajatgd | 23 July 2017, 21:05:34 UTC | adding test for Denoise Layer Propagation | 16 November 2017, 10:50:00 UTC |
| cd293fa | ajatgd | 23 July 2017, 21:01:52 UTC | Adding DenoisePropagation methods for Reference Architecture | 16 November 2017, 10:50:00 UTC |
| c5e4227 | ajatgd | 22 July 2017, 09:56:03 UTC | Fixing a small bug in Denoise Layer | 16 November 2017, 10:50:00 UTC |
| 2f7757b | ajatgd | 22 July 2017, 09:48:07 UTC | Adding Tensor input and Forward in Denoise Layer | 16 November 2017, 10:50:00 UTC |
| e825c01 | ajatgd | 21 July 2017, 10:42:57 UTC | Adding Transform Layer for Deep AutoEncoders | 16 November 2017, 10:50:00 UTC |
| 9c4ec94 | ajatgd | 21 July 2017, 08:59:50 UTC | Adding Denoise Layer for DeepAutoEncoders | 16 November 2017, 10:50:00 UTC |
| 1cc9168 | Saurav Shekhar | 21 July 2017, 00:27:28 UTC | ScaleAdd and GetMatrix functions on vectors added | 16 November 2017, 10:50:00 UTC |
| c5c820d | Saurav Shekhar | 20 July 2017, 19:11:45 UTC | RNNLayer added v1 | 16 November 2017, 10:50:00 UTC |
| 3c75a7d | IlievskiV | 20 July 2017, 09:19:47 UTC | test: Implement Conv Backpropagation test Implemented backpropagation test, the code is still buggy, fix commits will follow. | 16 November 2017, 10:50:00 UTC |
| 596261a | IlievskiV | 19 July 2017, 23:18:17 UTC | test: Implement Conv Prediction function test | 16 November 2017, 10:50:00 UTC |
| a9b1227 | IlievskiV | 19 July 2017, 22:55:43 UTC | test: Implement Conv Loss function test | 16 November 2017, 10:50:00 UTC |
| a35bbeb | IlievskiV | 19 July 2017, 21:37:26 UTC | test: Implement Forward pass test Construction of one simple conv net and forwarding a random matrix. | 16 November 2017, 10:50:00 UTC |
| 9786a41 | IlievskiV | 19 July 2017, 20:57:39 UTC | test: Implement function for creating test conv net | 16 November 2017, 10:49:59 UTC |
| 939895a | IlievskiV | 19 July 2017, 20:52:44 UTC | test: Add Im2Col, Downsample and RotateWeights tests | 16 November 2017, 10:49:59 UTC |
| 9d6d79c | IlievskiV | 19 July 2017, 12:15:02 UTC | feat: Implement Forward and Backward pass in Reshape Layer Implementation of the Forward and Backward pass in the Reshape Layer, which transforms the input to the desired output dimensions. | 16 November 2017, 10:49:59 UTC |
| f706c9d | IlievskiV | 19 July 2017, 12:13:05 UTC | feat: Define Reshape kernel Define the Reshape kernel for GPU and CPU architectures, implement it for the Reference architecture. | 16 November 2017, 10:49:59 UTC |
| a63e1b7 | IlievskiV | 18 July 2017, 12:38:57 UTC | fix: Include headers in Method DL | 16 November 2017, 10:49:59 UTC |
| 26f7ebf | IlievskiV | 18 July 2017, 12:38:30 UTC | fix: Change the method signatures | 16 November 2017, 10:49:59 UTC |
| b331a1f | IlievskiV | 18 July 2017, 12:37:37 UTC | fix: Wrong method names | 16 November 2017, 10:49:59 UTC |
| a38d2eb | IlievskiV | 18 July 2017, 12:33:09 UTC | feat: Add weight matrix in the Tensor Batch class Add the support for weighting each example in the batch. | 16 November 2017, 10:49:59 UTC |
| 1108346 | IlievskiV | 18 July 2017, 12:29:36 UTC | feat:Implement Deep Net class Implemen tation of the Deep Net class, which encapsulates everything for one deep neural network. | 16 November 2017, 10:49:59 UTC |
