jacobian

commit dac772bf6ffacef14787fc5f3e028dd6f9f4a169
parent a7bf9066c7cac609135db72e3e64bfaefe5ac26e
Author: David Freifeld <freifeld.david@gmail.com>
Date:   Tue, 30 Jun 2020 19:16:59 -0700

Working more w/ CNNs

Diffstat:
M
cnn.cpp
 | 
45
++++++++++++++++++++++++++++++++++++++++++++-
M
example.py
 | 
3
++-

2 files changed, 46 insertions(+), 2 deletions(-)
diff --git a/cnn.cpp b/cnn.cpp
@@ -1,26 +1,69 @@
 #include "bpnn.hpp"
 #include "utils.hpp"
 
+#define LARGE_NUM = 1000000 // Remove me.
+
 class ConvLayer
 {
   Eigen::MatrixXd* input;
   Eigen::MatrixXd* kernel;
   Eigen::MatrixXd* output;
+
+public:
+  void convolute();
 };
+
+void ConvLayer::convolute()
+{
+  for (int i = 0; i < input->cols() - kernel->cols() + 1; i++) {
+    for (int j = 0; j < input->rows() - kernel->rows() + 1; j++) {
+      output(j, i) = (*kernel * input->block<kernel->rows(), kernel->cols()(j, i)).sum()
+    }
+  }
+}
+
 class PoolingLayer
 {
+  Eigen::MatrixXd* input;
+  Eigen::MatrixXd* kernel;
+  Eigen::MatrixXd* output;
+
+public:
+  void pool();
 };
 
+void PoolingLayer::pool()
+{
+  // It doesn't look like anything better than O(n^4) is doable for this as kernel needs to go through matrix and you need to index kernel. LOOK INTO ME!! 
+  float maxnum = -LARGE_NUM;
+  for (int i = 0; i < input->cols() - kernel->cols() + 1; i++) {
+    for (int j = 0; j < input->rows() - kernel->rows() + 1; j++) {
+      for (int k = 0; k < kernel->cols(); k++) {
+        for (int l = 0; l < kernel->rows(); l++) {
+          if ((input->block<kernel->rows(), kernel->cols()(j, i))(l, k) > maxnum) {
+            maxnum = (input->block<kernel->rows(), kernel->cols()(j, i))(l, k);
+          }
+        }
+      }
+    }
+  }
+}
+
 class ConvNet : Network
 {
+  int stride_len;
+                
+  std::vector<ConvLayer> conv_layers;
+  std::vector<PoolingLayer> pool_layers;
 public:
   ConvNet(char* path, int batch_sz, float learn_rate, float bias_rate);
 };
 
-ConvNet::ConvNet(char* path, int batch_sz, float learn_rate, float bias_rate)
+ConvNet::ConvNet(char* path, int batch_sz, float learn_rate, float bias_rate, int stride)
 {
   learning_rate = learn_rate;
   bias_lr = bias_rate;
+  stride_len = stride;
   instances = prep_file(path, "./shuffled.txt");
   length = 0;
   t = 0;
diff --git a/example.py b/example.py
@@ -11,12 +11,13 @@ def bench(batch_sz, layers):
         net.add_layer(5, "relu")
     net.add_layer(1, "resig")
     net.initialize()
+#    net.list_net()
     net.train(50)
     end = time.time()
     return (end-init)
 
 timesum=0
-trials = 5
+trials = 20
 for i in range(trials):
     timesum+=bench(10,1)
 print("Averages over %s trials\n--------------\nTime: %s seconds.\n" % (trials, timesum/trials))