Initial checkin.

diff --git a/Makefile b/Makefile
new file mode 100644 (file)
index 0000000..1e14781
--- /dev/null
+++ b/Makefile
@@ -0,0 +1,9 @@
+CXX=g++
+CXXFLAGS=-O2 -ffast-math -g -Wno-write-strings $(shell pkg-config eigen3 --cflags)
+LDFLAGS=-lceres_shared
+
+OBJS=opt.o
+
+opt: $(OBJS)
+       $(CXX) -o $@ $(OBJS) $(LDFLAGS)
+

diff --git a/fit.sh b/fit.sh
new file mode 100755 (executable)
index 0000000..f3062d3
--- /dev/null
+++ b/fit.sh
@@ -0,0 +1,5 @@
+#! /bin/sh
+make
+wget -q -O- http://www.google.com/intl/en_ALL/ipv6/statistics/data/adoption.js | grep '^  \[' | cut -d, -f5 | cut -d\] -f1 > nativeadoption.dat 
+./opt nativeadoption.dat 
+

diff --git a/opt.cc b/opt.cc
new file mode 100644 (file)
index 0000000..20e6a61
--- /dev/null
+++ b/opt.cc
@@ -0,0 +1,81 @@
+// Optimizes general sigmoid function 1/(1 + exp(-a(t-m))) over a set of points.
+
+#include <stdio.h>
+#include <vector>
+#include "ceres/ceres.h"
+#include "gflags/gflags.h"
+#include "glog/logging.h"
+
+using namespace ceres;
+using std::vector;
+
+struct Point {
+       float t;
+       float adoption_rate;
+};
+
+class DistanceFromSigmoidCostFunction {
+ public:
+       DistanceFromSigmoidCostFunction(Point p) : p_(p) {}
+       
+       template<class T>
+       bool operator() (const T * const a, const T * const m, T* e) const {
+               *e = T(p_.adoption_rate) - T(1.0) / (T(1.0) + ceres::exp(- *a * (T(p_.t) - *m)));
+               return true;
+       }
+
+ private:
+       Point p_;
+};
+
+int main(int argc, char** argv) {
+       google::ParseCommandLineFlags(&argc, &argv, true);
+       google::InitGoogleLogging(argv[0]);
+
+       Problem problem;
+
+       FILE *fp = fopen(argv[1], "r");
+       if (fp == NULL) {
+               perror(argv[1]);
+               exit(1);
+       }
+
+       // The two parameters to be optimized.
+       double a = 1.0;
+       double m = 0.5;
+
+       int line_num = 0;
+       while (!feof(fp)) {
+               char buf[256];
+               if (fgets(buf, 256, fp) == NULL) {
+                       break;
+               }
+               Point p;
+               p.t = line_num;
+               p.adoption_rate = atof(buf) * 0.01;
+               problem.AddResidualBlock(
+                   new AutoDiffCostFunction<DistanceFromSigmoidCostFunction, 1, 1, 1>(
+                       new DistanceFromSigmoidCostFunction(p)),
+                   NULL,
+                   &a, &m);
+               ++line_num;
+       }
+       fclose(fp);
+
+       a = 1.0 / line_num;
+
+       // Run the solver!
+       Solver::Options options;
+       options.max_num_iterations = 1000000;
+       options.linear_solver_type = ceres::DENSE_QR;
+       options.minimizer_progress_to_stdout = true;
+
+       Solver::Summary summary;
+       Solve(options, &problem, &summary);
+
+       std::cout << summary.BriefReport() << "\n";
+       std::cout << "a=" << a << std::endl;
+       std::cout << "m=" << m << std::endl;
+       
+       return 0;
+}