search

opencog / opencog

A framework for integrated Artificial Intelligence & Artificial General Intelligence (AGI)
http://wiki.opencog.org/w/Development
Other
2.34k stars 723 forks source link

Demo ? #3398

Open freedomtrain opened 5 years ago

freedomtrain commented 5 years ago

Is there any documentation that gives instructions on how to create a demo of Opencog ?

Thanks

vsbogd commented 5 years ago

@fredhampton, could you please add details to your question. What is the goal of demo, target audience?

You can find materials of the OpenCog workshop at DevOps 2018 at https://dev.singularitynet.io/workshops/docker-opencog/ . Please contact @noskill, if you have questions on this materials.

freedomtrain commented 5 years ago

@noskill I have set up the docker demo with the instructions from this page. https://dev.singularitynet.io/workshops/docker-opencog/#running-the-vqa-demo

When I open the notebook and go to interface-images-demo there are no instructions ? Is there documentation that explains what the next steps are?

When I step through the code on the page and arrives at the block below I receive error messages.

models = os.path.expanduser('~/projects/data/visual_genome/') network_runner.runner = SplitMultidnnRunner(models) vqa = PatternMatcherVqaPipeline(extractor, question_converter, atomspace, None)

The error message is as follows:

loading dictionary from /home/relex/projects/data/visual_genome/dictionary.pkl no threshold for 25, using mean value 0.8608333333333336 no threshold for 90, using mean value 0.8608333333333336 no threshold for 107, using mean value 0.8608333333333336 no threshold for 114, using mean value 0.8608333333333336 no threshold for 180, using mean value 0.8608333333333336 no threshold for 220, using mean value 0.8608333333333336 no threshold for 225, using mean value 0.8608333333333336 no threshold for 285, using mean value 0.8608333333333336 no threshold for 392, using mean value 0.8608333333333336 no threshold for 613, using mean value 0.8608333333333336 no threshold for 630, using mean value 0.8608333333333336 no threshold for 631, using mean value 0.8608333333333336

noskill commented 5 years ago

Hi @fredhampton! It's is not error message, it is warning. Since it surprises users i am going to hide it.

Here this message says that some models doesn't have threshold. This demo uses model trained on balanced dataset. Coco vqa dataset we used for this demo(https://visualqa.org/download.html) is not properly balanced, so better performance may be achieved by fine-tuning model for this particular dataset. For balanced dataset threshold is 0.5, and here we shift it a bit.