On 03/18/2016 10:13 PM, Phil Bouchard wrote:

On 03/18/2016 09:49 PM, Peter Dimov wrote:

...

Phil Bouchard wrote:

...

Hi,

I wrote a small utility that reads PDFs over the Internet and converts the text into a tree of regular expressions. The goal here is to test root_ptr<> and see if it will stand up to this complexity.

The code is very simple for the moment as you can see: https://github.com/philippeb8/root_ptr/blob/master/example/t100_test1.cpp

I'm probably missing something... but isn't this kind of use enabled much more easily by using list<node> for the root and ordinary node* for the internal pointers? Once the list is destroyed everything is swept away, no need for smart pointers of any kind?

neuron_base got sub lists that will need to be explicitly destroyed as well in this case.

I will make it even more complex & cyclic tomorrow.

In implementing a search functionality I realized that there is no easy way to make a node_ptr<> from this. In my example I am forced to create a copy of *this so that I can create a note_ptr<> to it: node_ptr search(std::string const & input) { std::string res; boost::match_resultsstd::string::const_iterator what; if (boost::regex_match(input, what, exp_, boost::match_default | boost::match_partial)) if (what[0].matched) for (unsigned k = 1; k < what.size(); ++ k) if (what[k].matched) if (sub_.size() <= 1) return node_ptr(x_, new node(* this)); //< *** Copy of *this else for (std::list >::const_iterator i = sub_.begin(); i != sub_.end(); ++ i) for (std::list::const_iterator j = i->begin(); j != i->end(); ++ j) if (node_ptr p = search(input)) return p; return node_ptr(x_); } That's a caveat but the copy of *this is done quickly. Although search() could easily use neuron_base * because it is a read only function.

On 03/19/2016 01:53 PM, Phil Bouchard wrote:

return node_ptr(x_, new node(* this)); //< *** Copy of *this

It is always possible to create a new helper function: make_node_from_this(x_, * this);

On 03/18/2016 10:13 PM, Phil Bouchard wrote:

I will make it even more complex & cyclic tomorrow.

Its usage is pretty neat now: https://github.com/philippeb8/root_ptr/blob/master/example/t100.h#L73 root_ptr<> is pretty solid. I am not sure what else can possibly be done.

On 03/20/2016 01:35 PM, Phil Bouchard wrote:

It's really fun to program with.

Here I am returning complex r-values recursively and it works fine: https://github.com/philippeb8/root_ptr/blob/master/example/t100.h#L152

I've cleaned up the example again and now a simple: cout << "Mind dump:" << endl; cout << * t100 << endl; cout << "Searching for: \"einstein\"" << endl; if (node_ptr p = t100->search("einstein")) cout << p->sort().unique() << endl; cout << "Searching for: \"graviton\"" << endl; if (node_ptr p = t100->search("graviton")) cout << p->sort().unique() << endl; Will dump you the atomic sentence fragments related to your search.

Phil Bouchard wrote:

On 03/23/2016 10:48 AM, Phil Endecott wrote:

...

The subject line says "neural network", and although I see a few names like "neuron" in your code, I can't actually see a neural network implementation. It all seems to be about regular expressions.

What have I missed?

It's a general neural network with each neuron being represented by a regular expression. I said "tree" but it is really a graph of neurons randomly interconnected with each other: https://github.com/philippeb8/root_ptr/blob/master/example/t100.h#L72

It's not a "neural network" at all; it's a graph. Which might be what you need to test your smart pointer, but it's confusing to call it something that it isn't. Regards, Phil.