Darknet RNN Experiments

schwittlick commented 7 years ago

seems pretty fucking nice. except from it's written in C.

i've trained on 11mb of texts (80k lines) from all the parsted pdf's (125) and it took about 20h, finished with a loss of 0.4 and generated pretty nice shit.

The execution of the complexity of a computer is the scream .

The artist is essentially a considerable text .

Generative art , some expert is probably the connections between nature and contribution with an engineer , which were largely black onto an infinite meaning .

It is going on and he was going to have to stay in the future .

In fact , an objective digital art , the more we can function as a form of self-preservation .

In this sense , it is important to see the anti-subject : the dissolution of the Hypersphere of identity and communication systems .

It is part of the image between the audio power and sound .

But the determination of the algorithm is not so much a supposedly lived body .

The question generally promised the communication purpose .

it's possible to configure the generator:

-len <int>: change the length of text generated, default 1,000
-seed <string>: seed the RNN with the given string, default "\n"
-srand <int>: seed the random number generator, for repeatable runs
-temp <float>: set the temperature for sampling, default 0.7

one might have to make some adjustments to the code in order to parse the output of that program from an external python program. or write a parser for that output as it looks like that:

marcel@lyrik:~/devel/darknet⟫ ./darknet rnn generate cfg/rnn.cfg backup/all_texts_nov25.weights  -len 500 -temp 1.0
rnn
layer     filters    size              input                output
    0 RNN Layer: 256 inputs, 1024 outputs
                connected                             256  ->  1024
                connected                            1024  ->  1024
                connected                            1024  ->  1024
    1 RNN Layer: 1024 inputs, 1024 outputs
                connected                            1024  ->  1024
                connected                            1024  ->  1024
                connected                            1024  ->  1024
    2 RNN Layer: 1024 inputs, 1024 outputs
                connected                            1024  ->  1024
                connected                            1024  ->  1024
                connected                            1024  ->  1024
    3 connected                            1024  ->   256
    4 softmax                                         256
    5 cost                                            256
Loading weights from backup/all_texts_nov25.weights...Done!

No claim that information equal to the same hollow land beliefs that he had changed : for these issues of mediation every other error .
The telephone composers will it be in a strangered culture , this is not considerable through monopoly and other minds in which you could use it .
And , my source used for creativity is an interrelated study of network to the extraordinary powers of the area which is always to extend the epistemological determinism of a function of analysis .
And now found here

shouldn't be too hard though. the generation goes by pretty quick and takes about 1s for generating the above.

schwittlick commented 7 years ago

continued here https://github.com/mrzl/ECO/wiki/Darknet-RNN-Training-Documentation

Schwittleymani / ECO

Darknet RNN Experiments #156