edwinrobots / heatmapbcc Goto Github PK
View Code? Open in Web Editor NEWThis is a working version of the crowdscanner. Later commits have updated and improved the GP code but the crowdscanner demo may no longer look right.
This is a working version of the crowdscanner. Later commits have updated and improved the GP code but the crowdscanner demo may no longer look right.
Replace BCC with baysian classifer combination
Replace "situation" with enlarged gathering reports.
Remove numbers?
Crowd + Ensemble box can be simplified --> "to produce machine-readable data" is not required
Transition slide: highlight BCC box in the overview diagram
Mark into web page somehow so we can easily submit. E.g. when we submit first report, reset the lat and lon values to the new invalidating report.
send to nham for use by AO
Create a page in the web app that can display the current trust levels. Show confusion matrix histograms, an accuracy summary and an uncertainty value.
Has to be changed in both train() and post_grid() methods. Also need to alter the z value so it subtracts the correct mean if it is not 0.5.
Don't treat the t values as the observations, treat the updated kappa distribution as the observation. Q should match the variance in kappa, and the observation values should be the mean, taking into account nu0.
lower the radius of peak finder
the overview diagram is too much to take in during the demo -- separate demo and slide explanation.
possibly remove the overview diagram completely.
it's too loopy
could introduce it piece by piece, or only show some of the arrows when we actually discuss them
Compare for speed, accuracy (e.g. by looking at marginal likelihood with different parameters). Use GPy, but we'll need to submit the beta-adjusted values and have a way to input the corresponding observation noise.
It is hard to see the selected options/state in the checkboxes. Make these bold or make a new title to show if we are using Bayesian heatmaps
Do the maths to determine how we optimise for length scale as well. Can this be done as part of VB or as a ML2 step? It's important not to spend long on things that are not part of our core BCC model though.
also green background on prediction map is not great. Threshold so the 0.5 goes to 0?
Implement algorithm for finding peaks. Find all points with p(target) > theta. Eliminate those with neighbours that are higher.
Plot on map to test (as a binary grid). Use same method to display as for report intensity map.
Hook this up to GET requests by altering the demoserver.py
Needs to be done in both training and prediction.
For the demo, need to match types to those that the clients will want to display.
also ensure that final targets are in centre of city
Allow users to click on the map and pop up a list of ushahidi emergency reports that are close to the target points.
When the report ID is no longer included in the list, invalidate it by setting type in prov store to -1
Input length-scale is the important one, but also sigmoid sensitivity "s". Could either work out the marginal likelihood for HeatmapBCC, then use fminsearch. Alternativeily, there may be a way to handle the length-scale in an approximate Bayesian manner by including it in the variational updates using a delta method to approximate the non-conjugate posterior???
Other hyperparameters, i.e. alpha0 and nu0 (corresponds in latent space to output length scale and prior function mean) probably don't want optimising.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.