conveyal / analyst-server Goto Github PK

When you clock on PointSets, it says you can add a pointset and upload a shapefile. But that's not true, you have to upload a shapefile first.

It should probably not let you try to add a PointSet until you've uploaded a shapefile.

10 classes is too many; they can't be distinguished on the map.

Firefox 33, Ubuntu 14.04.1. When I go to 'add transport scenario' and fill in the form, clicking 'create' does nothing.

So that areas used for aggregation are not the same as pointsets.

If I create a scenario, wait for it to build and then change the name, I can't save it. I suspect this is because a property "status" gets set during graph building, which the server chokes on on save:

com.fasterxml.jackson.databind.exc.UnrecognizedPropertyException: Unrecognized field "status" (class models.Scenario), not marked as ignorable (9 known properties: "name", "processingGtfs", "projectId", "bounds", "processingOsm", "files", "id", "description", "failed"])
 at [Source: N/A; line: -1, column: -1] (through reference chain: models.Scenario["status"])
    at com.fasterxml.jackson.databind.exc.UnrecognizedPropertyException.from(UnrecognizedPropertyException.java:51)
    at com.fasterxml.jackson.databind.DeserializationContext.reportUnknownProperty(DeserializationContext.java:731)
    at com.fasterxml.jackson.databind.deser.std.StdDeserializer.handleUnknownProperty(StdDeserializer.java:915)
    at com.fasterxml.jackson.databind.deser.BeanDeserializerBase.handleUnknownProperty(BeanDeserializerBase.java:1298)
    at com.fasterxml.jackson.databind.deser.BeanDeserializerBase.handleUnknownVanilla(BeanDeserializerBase.java:1276)
    at com.fasterxml.jackson.databind.deser.BeanDeserializer.vanillaDeserialize(BeanDeserializer.java:243)
    at com.fasterxml.jackson.databind.deser.BeanDeserializer.deserialize(BeanDeserializer.java:118)
    at com.fasterxml.jackson.databind.ObjectMapper._readValue(ObjectMapper.java:3023)
    at com.fasterxml.jackson.databind.ObjectMapper.readValue(ObjectMapper.java:1637)
    at controllers.Api.updateScenario(Api.java:751)

If I give analyst-server 8GB with a Portland graph (doing a comparison), everything is fine. If I give it 4GB, the tiles silently fail to render. Also, it shouldn't need 8GB of RAM for Portland---OTP certainly doesn't need that much.

We need to build something (or find something someone else has built) with block-level datasets containing all Census data (LODES and decennial census) and tract-level (latest 5-year ACS). We can also use WorldPop data or LandScan for international coverage.

We say we're normalizing, but it's actually weighting. Normalization and group are both creating a weighted average, so they should be a single step. We need a slick way to generate reports.

For differences, we still want to show percentages of the total, rather than relative difference.

It takes a long time for the tiles and the histograms to come up. We should investigate why this is so slow.

There is no user feedback on creating a transit scenario until the file has transferred, which on a slow connection/large file can give the impression nothing is happening.

When I set the distance to 15 minutes I see no accessibility at the Census tract level:

With 30 minutes, it's fine:

I'm wondering if accessibility to self is not being considered (perhaps we're throwing out non-transit trips in profile routing?)

Data: https://www.dropbox.com/s/0dhtvh47daek0fx/acs_tracts.zip?dl=0

This requires some UI changes as well to allow the user to see min/max/average trip time.

The Jenks classifier is currently binning to intervals of 100. This may not be appropriate depending on dataset range, and also may create problems because it reduces the in-class variance, although @kpwebb points out that values are large enough this shouldn't matter.

For instance, when aggregating, it gets called on the disaggregate values and again on the aggregate values.

One of the Portland census tracts is coming up blank in aggregation:

It is somehow possible to get tiles that don't get cleared in regional mode. I will figure out how to reproduce.

Say what you're actually seeing.

When I create a new query the title always shows as blank.

When I upload a point-based shapefile:

• I can select any field (not just numeric fields)
• creating a new attribute spins forever.

Example shapefile: https://www.dropbox.com/s/5yq7c7bwt0zjikd/point_bug.zip?dl=0

The log messages are never written to the console, making debugging a challenge.

The colors in the legend don't match the colors on the map because the map tiles have opacity < 1. Probably the best we can do is to make the legend items partially transparent, and put a mpa-colored background behind them.

So that someone can check out the repo, type 'activator run,' and have data already.

The date should autoset to a reasonable default, but it's not.

See #39.

Currently, the request date is hardcoded in Analyst.java.

This will make it easier to see the difference.

When I kick off a batch job sometimes I see nothing but Vertex not found errors. This might be a graph passing issue.

The Census provides an internal point for all geographies; see the SF1 documentation, page 623.

When we group we are grouping by centroids not by overlaps. We should be allocating the appropriate percentage of the weight of each area to each larger unit of aggregation. We need to calculate the areas using spherical geometry.

Text is hard to read:

So it's clear how aggregation works.

During shapefile upload you see data-manager.shapefile-upload-description instead of the proper message text.

An exported shapefile should have all of the attributes that the imported one did.

The histogram is too sensitive to large jumps in numbers in a particular class (because a high-employment block was found in the 18 minute band, say). We should smooth it, because all the jobs in a block aren't concentrated at 18 minutes: some may 17, while some may be 19. Perhaps just bin by five-minute intervals on the client side?

STRtree appears to be thread-safe once built, so we can run multiple threads to do the calculations when aggregating.

And assign the weight pycnoplactically.

On large datasets, yeah. We could take the QGIS approach and take a random sample.