
Boost : 
From: Tom Brinkman (reportbase_at_[hidden])
Date: 20070722 13:31:46
boost::time_series is available here. The review stars at the end of the month.
http://boostconsulting.com/vault/index.php?directory=Math%20%20Numerics
Here are some initial questions and observations.
All questions in this post are related to the
simple_moving_average_functor, which
I have included at the bottom of this post.
I have more questions, but I thought this would be a good start.
1) A simple way to do a "simple_moving_average" is to use a std::deque,
adding elements using "push_back" and removing elements using "pop_front".
This technique can be used for any of the smoothing algorithms that
involve a fixed number of elements.
Do any of the boost::time_series containers mimic the "deque" capability to
add and remove elements at both ends of the array? I used a "deque" in
the sample "functor" at bottom of this post.
2) The "simple_moving_average_functor" has two functions,
"finalize()" and "operator()". (see below) The second function passes
in the offsets,
for "start" and "stop". However, the "finalize" function does not
pass any offsets.
Im not shur how to integrate these "offsets" into the calculating the
"moving average".
The functor that I created below just ignores them. Any suggestions?
3) After calling commit(), am I not able to add any more elements to a
"time_series"
container. In other words, do I have to recreate the "time_series"
container each time. Take for example, the case when Im calculating the
"simple_moving_average" over a period of 10 days. As I move through the data
set, all that I need to do is add another datatpoint, and then remove
the last data point.
To repopulate the "time_series" container each time would be nonoptimal. I
wonder if "time_series" container could release that data, and let me start
adding/removing elements to it again?
4) Do "range_run_storage::transform" and "range_run_storage::for_each" support
lambdas?
The example using "range_run_storage::transform" might look like this:
boost::range_run_storage::transform(series1, series2,
boost::lambda::_1 + boost::lambda::_2, inserter);
Example using "std::transform"
std::transform(series1.begin(),series1.end(),series2.begin(),
boost::lambda::_1 + boost::lambda::_2, std::back_inserter(series3));
5) What is the prefered method to do simple math against the elements in a
"time_series" container. Say for example I wanted to multiply the number
1.10 against each element. How would I do this using
boost::range_run_storage::for_each().
The example might look like this:
boost::range_run_storage::for_each(series1,
boost::lambda::_1 * boost::lambda::constant(1.10));
Example using "std::for_each"
std::for_each(series1.begin(),series1.end(),
boost::lambda::_1 * boost::lambda::constant(1.10));
template<typename out_t, typename series_t>
struct simple_moving_average_functor
{
std::size_t periods;
std::deque<double> data;
out_t out;
simple_moving_average_functor(const out_t& out, std::size_t periods,
const series_t& series) :
out(out), periods(periods){}
template <typename value_t, typename offset_t>
void operator ()(const value_t& value, offset_t start, offset_t stop)
{
if (data.size() < periods)
{
data.push_back(value);
return;
}
double sum = 0;
std::for_each(data.begin(),data.end(),
boost::lambda::var(sum) += boost::lambda::_1);
out(sum/periods, start,stop);
data.pop_front();
data.push_back(value);
}
out_t const &finalize()
{
double sum = 0;
std::for_each(data.begin(),data.end(),
boost::lambda::var(sum) += boost::lambda::_1);
return out(sum/periods);
}
};
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk