summary |
shortlog | log |
commit |
commitdiff |
tree
first ⋅ prev ⋅ next
jijoong.moon [Sun, 15 Mar 2020 23:40:27 +0000 (08:40 +0900)]
Make softmax optional
make softmax optional when the output is calculated
Resolves:
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Sun, 15 Mar 2020 23:02:03 +0000 (08:02 +0900)]
Update Bias when the optimizer is adam
Until now, Adam optimizer updates weight only. So that, adds Bias
update accordingly.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 13 Mar 2020 08:53:30 +0000 (17:53 +0900)]
Add Relu Activation Function
Add Relu Activation Function
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 13 Mar 2020 07:29:32 +0000 (16:29 +0900)]
Modify databuffer to generate random input variables at first time
Modify databuffer to generate random input variables at first time
- Add Conditional variable to synchronize thread
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 13 Mar 2020 07:28:08 +0000 (16:28 +0900)]
Fix the compilation Error for Batch Normalization
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 26 Feb 2020 04:46:03 +0000 (13:46 +0900)]
Batch Normalization Draft
- This is unstable version of Batch normalization
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 13 Mar 2020 04:33:47 +0000 (13:33 +0900)]
Add Keras & Tensorflow Example for transfer learning
mobilenetv2 output (features) using cifar10 data
: 10 Classes x 100 image each --> trainingSet.dat
* trainingSet.dat * ValSet.dat
62720 62720
+--------------+ +--------------+
| | | |
10x100 | | 10x10 | |
| | | |
| | | |
+--------------+ +--------------+
- Layers
: InputFeatures(62720)->10 hidden FC + softmax -> 10 Classes
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 26 Feb 2020 04:42:46 +0000 (13:42 +0900)]
Add Subtract with float value in tensor & add sum according to axis
Add Subtract(float value)
Add sum(int axis)
: sum according to axis
- 0 : batch
- 1 : height
- 2 : width
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 26 Feb 2020 04:39:45 +0000 (13:39 +0900)]
Change Configuration format fot batch normalization
Change Configuration format
- Delete Width, Hieght
- Add HiddenSize
Add BatchNormalizationLayer Scaleton Code
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 18 Feb 2020 08:14:48 +0000 (17:14 +0900)]
Add Softmax token in ini file
Softmax token is add to use to enable softmax at outputlayer
This is boolean and if it is set by true, then softmax is enabled.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 18 Feb 2020 07:47:28 +0000 (16:47 +0900)]
Fix calculation of Logistics Regression
Fix calculation of Logistics Regression at outputlayer
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Mon, 17 Feb 2020 11:49:16 +0000 (20:49 +0900)]
Add DataBuffer for big data
- Add DataBuffer Class to read/write big data from file
- Multi-Threaded
- Automatic Updated
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 13 Feb 2020 22:49:47 +0000 (07:49 +0900)]
Calculation of Validation Loss & Accuracy
Calculate Validation Loss & Accuracy
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 13 Feb 2020 07:37:05 +0000 (16:37 +0900)]
Add Classification Example (cifar10 data set)
Classification Example is added. Data Set is cifar10.
The configuraion is in Classification.ini.
The base model for feature extractor is mobilenet v2 with pretrained
imagenet data. ( 244x244x3 ) Input feature size is 1280x7x7 = 62720.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 13 Feb 2020 01:50:11 +0000 (10:50 +0900)]
Implement Cross Entropy Cost Function
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 12 Feb 2020 05:13:54 +0000 (14:13 +0900)]
Using CBLAS for Tensor Calculation
Implement Tensor Calculation for Tensor Calculation
Can use with "-DUSE_BLAS"
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 11 Feb 2020 06:05:39 +0000 (15:05 +0900)]
Decayed Learning Rate
Implement Decayed Learning Rate for better convergence
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 11 Feb 2020 02:12:42 +0000 (11:12 +0900)]
Fix Maintainer in README.md
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Mon, 10 Feb 2020 04:48:00 +0000 (13:48 +0900)]
Add README.md
README includes Descriptions of NNtrainer, how-to's and Open Source License.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Mon, 10 Feb 2020 01:36:13 +0000 (10:36 +0900)]
Use float data type instead of double.
For the efficiency, we are going to use float data type.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 7 Feb 2020 05:35:13 +0000 (14:35 +0900)]
add .gitmodules
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 7 Feb 2020 05:12:14 +0000 (14:12 +0900)]
Change Class Name Matrix to Tensor
Tensor is more appropreate than Matrix.
Resolves:
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 7 Feb 2020 04:49:28 +0000 (13:49 +0900)]
Code Refectoring
Code Refactoring
- Move to applicatoins into Application Directory
- Remove NeuralNet Directory
- Make include & src directory for the neural network
- Modifiy build configuration to build with libnntrainer.so
- Add nntrainer install & add pc.in for pkgconfig
- Add install path
**Changes proposed in this PR:**
- Added TOC generator for README.md
Resolves:
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 5 Feb 2020 08:04:33 +0000 (17:04 +0900)]
Make nntrainer shared library
Change Directory structure and make nntrainer shared library,
nntrainer.so
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Mon, 20 Jan 2020 01:07:01 +0000 (10:07 +0900)]
Add License file
Add Apache 2.0 full license file
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 5 Dec 2019 23:56:23 +0000 (08:56 +0900)]
Add Copyright for Appache 2.0 License
Add Copyright
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 4 Dec 2019 12:24:16 +0000 (21:24 +0900)]
add doxygen doc for DeepQ
add doxygen doc for DeepQ
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 4 Dec 2019 11:32:40 +0000 (20:32 +0900)]
add doxygen doc for KNN example
add doxygen doc for KNN example
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 4 Dec 2019 11:23:25 +0000 (20:23 +0900)]
add doxygen doc for Logistic regression
add doxygen doc for Logistic regression
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 4 Dec 2019 10:59:36 +0000 (19:59 +0900)]
add doxygen document for Training
add doxygen documents for Training
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 4 Dec 2019 10:16:53 +0000 (19:16 +0900)]
Add Doxygen Documentation for Environment
Add Doxygen Doc. for Environment
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 4 Dec 2019 08:15:47 +0000 (17:15 +0900)]
Add doxygen Documentation for NeuralNet
Add doxygen Documentation format of NeuralNet Directory
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 26 Nov 2019 01:58:25 +0000 (10:58 +0900)]
fix to use epsilon from config file
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 26 Nov 2019 01:20:16 +0000 (10:20 +0900)]
Add Logistic Regression
Add Logistic Regression with example
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Mon, 25 Nov 2019 07:43:06 +0000 (16:43 +0900)]
add network description for generality
- include iniparser
- add parse configuratio ini file to describe network
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 22 Nov 2019 01:57:16 +0000 (10:57 +0900)]
Add bias correction of adam optimizer
add bias correction according to the epoch.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 21 Nov 2019 10:23:23 +0000 (19:23 +0900)]
Add CMakeLists.txt for KNN & support ubuntu
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 21 Nov 2019 09:54:24 +0000 (18:54 +0900)]
Add README for DeepQ
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 21 Nov 2019 09:26:26 +0000 (18:26 +0900)]
Add Layers::Layer classes
Add Layers
. Layers::Layer
. Layers::InputLayer : public Layer
. Layers::FullyConnectedLayer : public Layer
. Layers::OutputLayer : public Layer
with forwarding, backwarding support
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 20 Nov 2019 04:49:53 +0000 (13:49 +0900)]
Modify code for clang format
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 19 Nov 2019 22:48:27 +0000 (07:48 +0900)]
Restructure neural network
Make NeuralNet directory for common usages.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 19 Nov 2019 04:32:34 +0000 (13:32 +0900)]
Add Android.mk for android build
- Add Android.mk
- Remove Boost Dependency
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 19 Nov 2019 00:11:20 +0000 (09:11 +0900)]
Add Custom Environment Cartpole
Depending the USING_CUSTOM_ENV definition, can choose custom
env (Env::CartPole) or Gym::Environment.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 19 Nov 2019 00:05:00 +0000 (09:05 +0900)]
Add Custom Environment : CartPole
Implement Custom CartPole Environment with test example
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Mon, 18 Nov 2019 06:31:12 +0000 (15:31 +0900)]
add reference model.bin
Add Trained model for CartPole-v0 ( Terminate 200 Iteration Max )
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Mon, 18 Nov 2019 06:29:41 +0000 (15:29 +0900)]
add ADAM (Adaptive Momemt Estimation) to accelerate convering
- Add ADAM optimizer
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 15 Nov 2019 07:58:37 +0000 (16:58 +0900)]
Add Another Fullyconnected Layer
Add Fully Connnected Layer : ( Two FC )
Resolves:
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Fri, 15 Nov 2019 07:55:20 +0000 (16:55 +0900)]
Add tanh & minibach mode
- Add tanh activation
- Add minbatch mode
Resolves:
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 12 Nov 2019 04:42:55 +0000 (13:42 +0900)]
Implement DeepQ Learning without minibatch
Implement DeepQ Learning for CartPole-v0
Currently It is not stable.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 7 Nov 2019 00:29:52 +0000 (09:29 +0900)]
Add initialize DeepQ with gym (CartPole-v0)
Add DeepQ for CartPole-v0
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Wed, 6 Nov 2019 23:03:16 +0000 (08:03 +0900)]
Add NeuralNetwork Class
Add Network namespace & NeuralNetwork Class.
Add copy Matrix & NeuralNetwork
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 17 Oct 2019 06:44:09 +0000 (15:44 +0900)]
add readme
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Thu, 17 Oct 2019 06:37:07 +0000 (15:37 +0900)]
Add retraining example
Add retraining example. ( Forward and Backward Propagation for fully
connected layer )
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 15 Oct 2019 01:53:50 +0000 (10:53 +0900)]
Add README.md
Add some results in README.md
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
jijoong.moon [Tue, 15 Oct 2019 01:42:24 +0000 (10:42 +0900)]
Toy Example for the Transfer learning
Simple example of transfer learning like apple's Sticker example.
Mobilenet SSD V2 tensorflow lite model is used for the feature
extractor and KNN is used for the classification.
It is tesed on the android (Galaxy S8).
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
문지중/On-Device Lab(SR)/Principal Engineer/삼성전자 [Tue, 15 Oct 2019 01:09:34 +0000 (10:09 +0900)]
Initial commit
projects / platform / core / ml / nntrainer.git / log