RapidLib issueshttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues2017-08-30T12:15:17Zhttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/37Create IML API overview2017-08-30T12:15:17ZMichael ZbyszyńskiCreate IML API overviewOnce all of our stuff and ircam's stuff are together in a Node module ( #35 #36 ), write up a quick overview of the functions, overlaps, and conflicts.Once all of our stuff and ircam's stuff are together in a Node module ( #35 #36 ), write up a quick overview of the functions, overlaps, and conflicts.Future APIMichael ZbyszyńskiMichael Zbyszyńskihttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/46Decide which model types the API should support2017-05-02T14:07:05ZMichael ZbyszyńskiDecide which model types the API should support#45 and #44 exist for random forest and svm.
Which other model types should be part of the API.#45 and #44 exist for random forest and svm.
Which other model types should be part of the API.API v0.2https://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/86DTW: group examples into "Gesture types"2017-08-25T16:16:54ZMichael ZbyszyńskiDTW: group examples into "Gesture types"Wekinator lets users record multiple examples of a "Gesture Type". The current RapidLib implementation considers each example as a unique type. This matches how DTW works; the algorithm doesn't do any grouping.
Implement Gesture Types s...Wekinator lets users record multiple examples of a "Gesture Type". The current RapidLib implementation considers each example as a unique type. This matches how DTW works; the algorithm doesn't do any grouping.
Implement Gesture Types so RapidLib is more like Wekinator.Michael ZbyszyńskiMichael Zbyszyńskihttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/87DTW "normalisation"2017-08-30T11:52:39ZMichael ZbyszyńskiDTW "normalisation"This is probably more complex than can be addressed in RapidLib.
Some kinds of normalisation could be useful for certain DTW matching. Shifting a gesture in space and/or in size will increase the matching cost in DTW. Location and siz...This is probably more complex than can be addressed in RapidLib.
Some kinds of normalisation could be useful for certain DTW matching. Shifting a gesture in space and/or in size will increase the matching cost in DTW. Location and size could be normalized. For example, a user might want to match the letter "Z" no matter where it is drawn on a canvas, or whether it's big or small.Future APIhttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/61Wekinator-like output limits2017-08-16T13:17:22ZMichael ZbyszyńskiWekinator-like output limitsWekinator has some options for limiting the output. Duplicate those in our API.Wekinator has some options for limiting the output. Duplicate those in our API.Future APIhttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/40Set of use cases grounded on "off-the-shelf" sensors2018-02-13T18:06:14ZFrancisco BernardoSet of use cases grounded on "off-the-shelf" sensorsMyo,
Leap Motion,
Kinect,
Groove IoT packMyo,
Leap Motion,
Kinect,
Groove IoT packFrancisco BernardoFrancisco Bernardohttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/90RapidLib: Classifiers with string labels2017-08-30T12:05:37ZMichael ZbyszyńskiRapidLib: Classifiers with string labelsRapidLib classifiers should natively deal with strings as labels, rather than rely on the unordered_map in the facade.RapidLib classifiers should natively deal with strings as labels, rather than rely on the unordered_map in the facade.Future APIMichael ZbyszyńskiMichael Zbyszyńskihttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/93Minimum number of hidden nodes?2017-08-18T10:53:48ZMichael ZbyszyńskiMinimum number of hidden nodes?I've noticed that a one input neural network isn't very effective. It can't express a very complex curve, since it only has a series of two sigmoid nodes to work with.
Maybe the hidden layer should have a minimum number of nodes?I've noticed that a one input neural network isn't very effective. It can't express a very complex curve, since it only has a series of two sigmoid nodes to work with.
Maybe the hidden layer should have a minimum number of nodes?https://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/39Feature extraction pipeline2017-09-20T10:05:07ZMichael ZbyszyńskiFeature extraction pipelineIntegrate our IML into PiPo? Essentia as PiPo?Integrate our IML into PiPo? Essentia as PiPo?API v0.3https://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/30Code review of RapidLib against API Design Checklist2017-09-20T10:05:07ZFrancisco BernardoCode review of RapidLib against API Design ChecklistTest RAPIDLIB API Design against
https://theamiableapi.com/2012/01/16/java-api-design-checklist/Test RAPIDLIB API Design against
https://theamiableapi.com/2012/01/16/java-api-design-checklist/Francisco BernardoFrancisco Bernardohttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/106Thread hogging behaviour when training a classification/regression model2018-01-19T14:11:31ZFrancisco BernardoThread hogging behaviour when training a classification/regression modelThe "thread hogging" behaviour of the JS client API is problematic as it blocks the whole interface. This is a problem with a severity level between Major Usability problem and Usability catastrophe (Nielsen), observed with high frequenc...The "thread hogging" behaviour of the JS client API is problematic as it blocks the whole interface. This is a problem with a severity level between Major Usability problem and Usability catastrophe (Nielsen), observed with high frequency between different actions and users.
For instance, one participant, decided to move away from using the JS client library to trying to use a server side implementation (which also was problematic on its own). It would be beneficial to explore a Web-worker implementation to surpass this problem.
Recommendation:
Pursue a design pattern in which there are two learning modules (i.e., two classification modules, or two regression modules) one on the regular JS client code, another on the web-worker code. The web worker code receives training data from regular code, trains the model, and exports the JSON model to the regular JS code model, that loads it and runs test data with it.Future APIFrancisco BernardoFrancisco Bernardohttps://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/108API methods for training data quantification2018-01-19T14:25:20ZFrancisco BernardoAPI methods for training data quantificationThe quantity of data should be made visible on the high level interfaces — quantity of data can have multiple aspects to it — (e.g., #recorded Rounds, #examples per round, #total memory consumption). In one instance, one participant no...The quantity of data should be made visible on the high level interfaces — quantity of data can have multiple aspects to it — (e.g., #recorded Rounds, #examples per round, #total memory consumption). In one instance, one participant noticed the memory of the browser was reaching 3GB, and found this unusual, and also observed the consequences in training time.
Recommendation: Participants would benefit from API methods (namely in the training data class) that provide metrics about quantity or volume of data. This should help developers in design by making them explicit through the visual interface metaphors they develop. It should also be explained in the documentation the consequences of dealing with large data sets (e.g., training time, impact on the classification results outcomes, etc.)https://gitlab.doc.gold.ac.uk/rapid-mix/RapidLib/-/issues/68Implement namespace2017-11-08T16:50:01ZMichael ZbyszyńskiImplement namespaceWe should probably have a single level of namespace for our library
rapidmix::
rapidlib::
??We should probably have a single level of namespace for our library
rapidmix::
rapidlib::
??API v0.3