jacobian

commit 0e501e7d0e142ae992cb5d0b7576f06f25e7d673
parent ecc13627b7bc17e1c378d6055615a198dbc242d0
Author: David Freifeld <freifeld.david@gmail.com>
Date:   Thu, 30 Jul 2020 21:49:22 -0700

Tweaks + readme update

Diffstat:
M
readme.md
 | 
25
++++++++++---------------
M
src/cnn.cpp
 | 
16
+++++++++-------
M
tests/simd_simple.cpp
 | 
12
++++++------

3 files changed, 25 insertions(+), 28 deletions(-)
diff --git a/readme.md b/readme.md
@@ -63,21 +63,16 @@ Finally, train your network for one epoch with `train()`. Training one epoch at 
 ### Examples
 In the `/scripts` directory there is a example of a neural network being used in conjuction with Weights & Biases, allowing for effective hyperparameter searches and accuracy reporting.
 
-## Compiling/Installing
-
-### Precompiled Libraries
-**Coming soon:**: Precompiled pyc files and binaries in 'Releases' section for you if you don't want to build from source.
-
-### Python Bindings
-
-1. Install both the C++ end of pybind11 and the python end.
-2. Build with your chosen configuration using `make`.
-3. Copy the `mrbpnn.cpython-37m-darwin.so` file into your personal project directory.
-4. Import `mrbpnn` from your Python code and use it.
-5. Look to the `example.py` file for simple usage of the library.
-
-**Coming soon:** A guide on how to properly install Jacobian as a python library.
-**Coming soon:** Availability through package managers or a less convoluted install process.
+## Installing
+
+### Main Steps
+Note: This is all ideally the process, but it's so confusing that I'm not sure. You're better off manually copying the .so file!
+1. Install the C++ library Eigen.
+2. Install the C++ and python ends of the pybind11 library.
+3. Run `make` with the configuration of your choosing.
+4. Run `python setup.py bdist_wheel`
+5. `cd` into `dist` and run `python3 -m pip install --upgrade Jacobian-1.0-cp37-cp37m-macosx_10_13_x86_64.whl`
+6. Be unhappy when it doesn't work out and resort to just copying .so files.
 
 ### Build Configurations (Building from Source)
 
diff --git a/src/cnn.cpp b/src/cnn.cpp
@@ -8,6 +8,8 @@
 #include "bpnn.hpp"
 #include "utils.hpp"
 
+#include <Eigen/unsupported/CXX11/Tensor>
+
 #define LARGE_NUM 1000000 // Remove me.
 
 #if (!RECKLESS)
@@ -250,13 +252,13 @@ void ConvNet::process()
   // Assumes pooling is immediately after any conv layer.
   for (int i = 0; i < preprocess_length-1; i++) {
     conv_layers[i].convolute();
-    pool_layers[i].input = conv_layers[i].output;
-    pool_layers[i].pool();
+    //   pool_layers[i].input = conv_layers[i].output;
+    //   pool_layers[i].pool();
     conv_layers[i+1].input = conv_layers[i].output;
   }
   conv_layers[preprocess_length-1].convolute(); 
-  pool_layers[preprocess_length-1].input = conv_layers[preprocess_length-1].output;
-  pool_layers[preprocess_length-1].pool();
+  //pool_layers[preprocess_length-1].input = conv_layers[preprocess_length-1].output;
+  //pool_layers[preprocess_length-1].pool();
   //  std::cout << "Output:\n" << *pool_layers[preprocess_length-1].output << "\n\n";
   Eigen::Map<Eigen::RowVectorXf> flattened (conv_layers[preprocess_length-1].output->data(), conv_layers[preprocess_length-1].output->size());
   // std::cout << "Flattened:\n" << flattened << "\n\n";
@@ -379,10 +381,10 @@ int main()
   ConvNet net ("../data_banknote_authentication.txt", 0.05, 0.01, 5, 0.9);
   Eigen::MatrixXf labels (1,1);
   net.add_conv_layer(28,28,1,9,9,0);
-  net.add_pool_layer(20,20,1,6,6,0);
+  //  net.add_pool_layer(20,20,1,6,6,0);
   net.add_conv_layer(15,15,1,6,6,0);
-  net.add_pool_layer(10,10,1,2,2,0);
-  net.add_layer(81, "sigmoid");
+  //net.add_pool_layer(10,10,1,2,2,0);
+  net.add_layer(400, "sigmoid");
   net.add_layer(5, "lecun_tanh");
   net.add_layer(10, "resig");
   //  net.list_net();
diff --git a/tests/simd_simple.cpp b/tests/simd_simple.cpp
@@ -13,23 +13,23 @@ int main()
   Eigen::MatrixXf a = Eigen::MatrixXf::Random(1, sz);
   Eigen::MatrixXf::Index maxRow, maxCol;
   float max = a.maxCoeff(&maxRow, &maxCol);
-  Eigen::MatrixXf m1 = ((a.array() - max) / ((a.array() - max).exp().sum())).matrix();
+  Eigen::MatrixXf m1 = ((a.array() - max).exp() / ((a.array() - max).exp().sum())).matrix();
   auto softmax_simd_start = std::chrono::high_resolution_clock::now();
   Eigen::MatrixXf a1 = (1 + (-1 * a.array()).exp()).pow(-1).matrix();
   auto softmax_simd_end = std::chrono::high_resolution_clock::now();
   auto softmax_start = std::chrono::high_resolution_clock::now();
-  float sum = 0;
+  // float sum = 0;
   Eigen::MatrixXf::Index maxRow2, maxCol2;
   float max2 = a.maxCoeff(&maxRow2, &maxCol2);
   Eigen::MatrixXf m2 = (a.array() - max2).matrix();
-  for (int j = 0; j < m2.cols(); j++) {
-    sum += exp(m2(0,j));
-  }
+  // for (int j = 0; j < m2.cols(); j++) {
+  //   sum += exp(m2(0,j));
+  // }
+  float sum = ((a.array() - max2).exp().sum());
   for (int j = 0; j < m2.cols(); j++) {
     m2(0,j) = exp(m2(0,j))/sum;
   }
   auto softmax_end = std::chrono::high_resolution_clock::now();
-  std::cout << m1 << " | " << m1.sum() << "\n\n" << m2 << " | " << m2.sum() << "\n\n";
   if (m1.isApprox(m2)) std::cout << "Softmax matrices are equal!\n\n";
 
   double eigen_time = std::chrono::duration_cast<std::chrono::nanoseconds>(softmax_simd_end - softmax_simd_start).count();