Harry Plotter - Tcp Ip send recieve data Graphing

Harry Plotter is the upgraded graphing program that was originally used for my RBF project and has now been upgraded to using Tcp/Ip data streams.

If you have a server running your AI programs making use of a GPU for faster processing and you have like me a little netbook which has about 1 core the it is usefull to have a Harry Plotter.

The code to create a Harry Plotter needs a tidy but its really simple:

A class for creating an X11 window and Graph plotter and a class for the Tcp protocol. I will upload the code once it is tidy.

But here is an upload of the binary's. You need pthread lib, and X11 lib's and then it should run on Linux.

Here's a screen shot of Harry Plotter. If you get it working then email me for the source code.

Usage:
 Open a tcp port on your server and laptop using command:
sudo ufw allow from 'ipaddress' to any port 'port'

On your laptop / Receiver
./tcp_rcv_graph 'port'


On your Server / Transmitter
./tcp_snd 'ipaddress' 'port'

e.g Ipaddress is the address of the machine with the receiver.

Now you can send your data across your network or even across the internet.

Here is the Tcp/Ip Receiver Source:
tcp_rcv.cc

Here is the Tcp/Ip Sender Source:
tcp_snd.cc

Here are the Binarys for HarryPlotter:
tcp_rcv_graph        and      tcp_snd

Please Post a comment to get the Complete Source code for Harry Plotter.

Hope that was helpful.

Stream Amazon Prime over your Ubuntu Linux network

Wow so you've decided to stream your music now and MPD is useless
Amazon prime has all your albums and its affordable but so far you can only
listen via your browser or phone. You want it streamed throughout your
home network to every machine you own just like mpd.

Heres how to do it using vlc :
1. Open a browser on your Ubuntu box and start streaming music to your soundcard.

2. Issue this command on your Ubuntu box to find out what sound device to use:
pacmd list-sources | awk '/name:.+\.monitor/'

3.  Then add this information to the following cvlc command:

cvlc pulse://alsa_output.pci-0000_00_1b.0.analog-stereo.monitor --sout '#standard{access=http,mux=ogg,dst=you.ripa.dre.ss:8080}'

4. Open http://you.ripa.dre.ss:8080 on any of your other machines to stream
directly from your amazon prime music! Yeah.

And it works and now you can listen your Music just like with MPD.

Seed Recall

Ok so its time to fess up about what Ive discovered. OMG. Seed recall is a discovery I made whilst experimenting with the RLnet - Neural Q Learner.

I noticed that the solution depened on the random seed. So I decided to test the theory that the randomly perturbed activation functions themselves encoded different solutions (and heres the big one) for the same prelearnt weights.

Heres what I did:

Learning:
seed random generator
init random weights or load previous weights
create seed during learning - using Random Decay (See previous post)
backpropogate solution 178 iterations
save seed
save weights

repeat for all solutions

Recall:
set random generator to null
load saved weights
load seed for solution - Apply seed directly to Activation function
backpropogate        58 iterations
view seed dependant solution

For each seed I got a unique solution. Ofcourse its early days but it may just be that different solutions to the same problem can be learnt and encoded by the same neural network using the different random seeds applied to the activation function for each neuron.

Whether this extends to other problems I have yet to discover. If the problems are similar and have the same dimensions then it is possible.

The problem I am using is that detailed in previous posts:

{{-1, -1, -1,-1,  0,-1},
            {-1, -1, -1, 0, -1, 1},
            {-1, -1, -1, 0, -1,-1},
            {-1,  0, -1,-1,  0,-1},
            {-1, -1, -1, 0, -1, 1},
            {-1,  0, -1,-1,  0, 1}};
6 States with 6 Actions

I now plan to Upscale the RLnet to do 120x120 and use a CNN layer.

Cumin Tikka

Invented new dish calling it CuminTikka. Marinade chicken in yoghurt, cumin,corriander,paprika,garam masala, ginger and garlic. Overnight.
Roast chicken marinades.( Or You can just add the Marinades without cooking them)
Add cumin 2tsp, freshly chopped mint to onions cook down.
Brown blanched potatoes with butter,herbs and garlic.
Add tomatoe to onions, add stock, and lemon juice.
Add all to sauce and then to casserole.
It is very refreshing the cumin, mint and lemon sauce is delicious and fresh, the marinades are spicey and fun.
CuminTikka.

You can roast the marinades and then add them, or add them without roasting or frying. The flavour locks into the chicken from the marinade. Even with adding marinade to the sauce the lovely fresh taste of the cumin-mint still comes through.

Finally the Quest for the Holy grail of (My)AI is nearly over...

Since 2007 I had begun a quest to train a neural network to learn
the Bellman Residual - the equation used in Reinforcement Learning (Q learning).

As the years went by Deep Learning took over the quest and completed it (and some!!).

However the Quest was still ellusive to me and so I went off and did something else with my spare time.

But recently I returned to the quest. I dry docked an MLP neural network and benchmarked it on Xor. Then I set about applying it to the problem specified here:

 {{-1, -1, -1, -1, 0, -1},
            {-1, -1, -1, 0, -1, 1},
            {-1, -1, -1, 0, -1, -1},
            {-1, 0, -1, -1, 0, -1},
            {-1, -1, -1, 0, -1, 1},
            {-1, 0, -1, -1, 0, 1}};

Each of the 6 array above is a state with 6 actions. It is a Qlearning problem used by Mnemstudio.

So as you may have read in a previous blog I tried using Mnemstudio's XOR Backpropogating Neural Network. But alas I knew it needed another layer.

I tried adding another hidden layer, a CNN layer all to no avail. Until I rewrote it completely and applied my own learning rules.

I also cheated and used a RANDOM DECAY routine that decreased with Error. That means that while it learnt its TRANSFER FUNCTION was slightly random meaning that it could jump out of SUBOPTIMAL SOLUTIONS and find new ones.

With 4 layers : INPUT 6 X NEURONS WITH NORMALISED DATA
                        OUTPUT 6 X NEURONS WITH SIGMOID ACTIVATIONS
HIDDEN LAYERS X 2 WITH 18 NEURONS WITH TANH AND SIGMOID

For this algo the CNN option doesnt help for this problem however I am hoping it will work on Pixel to action problems. (Where the input is a image and the output is an action)

What was interesting is that the neural network converges to different solutions where as the Q learner stays in one solution - which I think is the purpose of using a Neural Network for RL. Its power is that it can find solutions where the Q learner cannot.

Q test output
1, 5
3, 4, 5
5, 5
2, 3, 4, 5
4, 5
0, 4, 5
Q Neural test output
1, 5
3, 1, 5
5, 4, 5
2, 3, 1, 5
4, 5
0, 4, 5

Two different solutions above both equally valid.

How did we get an error curve like that!!!

Sneak a peak @ Shepherds latest works DarkSide - Chapter 2 - The Watchers


Or Read Kromos now

Space Builders WANTED

Which wealthy nation with a history of empire will be the first to grasp this amazing opportunity to begin building its own fleet of spaceship's in orbit around earth. Musks ambitious and genius accomplishments
have furthered the space race - a reliable falcon 9 reusable rocket that can inject people into orbit via its dragon spaceship and the falcon heavey to transport tonnes of materials into orbit (also reusable) meaning that any country could begin building an emmense spacefleet now to rival those of any other country.

The Women of Mars - Prt 1 Growing Women on Mars

Growing women on mars.

The spaceship aurora entered its docking station just above fort henry. Isobel watched the greyish red planet opening out beneath her its red halo like circumference was eclipsed by the gateway staion that orbited above the martian sprawl beneath that opened out like a web of interconnected habitat's covering the equilivent land mass of the united states. Half above half below just one continuos city all made from martian soil and minerals. As she readied herself to jump from the spacecraft she held the spaceship one last time after all it had been her home for more than a year. The journey from Europa had been long and she was very tired. She would have liked to have been on the Earth shuttle it took only 3 weeks but that would have meant growing up on earth and that was only in her dreams no one from the outer planets could visit earth due to gravity and population controls.

It was generally well known that child birth by an earth woman on mars was impossible due to the lack of gravity. This had caused many problems for the colonists many of whom were family's and to discover that all births were medically impossible was a serious stumbling block for the new world. On Europa there wasnt a problem as gravity on board the Jupitor zone stations had been controlled to keep it at the right level for fetal development. On Europa if you had become impregnated then you had to make the routine trip to one of the many prenatal way stations in the Jupitor system. This often meant the parents forking out for either an abortion or paying for her stay 
upstairs until she gave birth. As gravity interferred with fetal development it was generally advised that no more than a few weeks should be left before relocation. On Mars however this wasnt possible the gravity stations around Mars were underequipped and many didnt have gravity control. The reason being – Mars was a poor planet. It was inhabited mostly by pioneers and renegades who had become obsessed with terraforming and sunk all of their money in turning Mars into earth. The gateway was all they could afford it was simply a jumping off point for arrivals who werent able to afford the earth shuttle and didnt mind free falling to a dead planet. What they saved on freefall they could use to pay for the return journey and the expensive orbital lifter which often broke down.

So unlike the Jupitor system and ofcourse earth on Mars baby's were a rarity. And no baby's meant fewer women. So most of the women came over from Earth or Jupitor attracted to the Martian experience and the Men of Mars. Who for the most part were a hardy bunch.

Mars sanitation and health department was inundated with problems that stemmed from Martian Men impregnanting their women and then the failed preganancy's and all of that mucky stuff. So they decided to make a woman from scratch that would satisfy the men on mars and be able to produce offspring.

They believed if done right they wouldnt know the difference. They were wrong. The hybrid women were unbelievably stunning they had tall athletic bodys and extremely good looks. The scientists at Mars health department couldnt understand why they were all such stunners. “Something to do with the gravity dna switch.”

Blurb :

Isobella a no nonsense girl from way out west sorts out the men of mars who dont no what to do inundated with all these femme fetals and helps the femme fetals create their own ideal men using the Mars tech for cloning hybrid humans she then goes of in search of the one man that she might faguely fancy as all the martian men fancy her but she's just not interested and heres about thisw guy whose living way out at the poles. This guy was one of the earliest pioneers before all the gruff types turned up and was obsessed with terraforming mars using some advanced technology instead he develops the ability to impregnate Isobella with this technology whose purpose was to start life on mars. Now pregnant and not sure how Isobella returns angrily to the space port to prepare for departure back to Jupitor. The shuttle fails and falls back to the planet and she is jettisoned but is rescued by a tribe of hybrid mars women who hail her as their savour. She gives birth to a normal healthy girl who is the first true woman of mars.

The Hybrid mars women have special lungs to allow them to breath on mars hence live outside and have resistance to the cold. They are also extremely beautiful and have telepathic powers to seduce any man or woman. And babies loads and loads of babies. The end.

Read More Go to juice.extramindcorp.com
Buy the book  'Seed Army' at Amazon.