Improving invalid-tag-modification comparator

[bkowshik]

Per stats here: #143 (comment) and continuing from here: #127

---

Invalid tag modification labels a lot of changesets, how can we improve this comparator?

Total stats

• Changesets labelled "Invalid tag modification": 7,644
• Changesets checked: 220
• Changesets harmful: 25
• Hit rate: 11.36%

Daily stats

• Average changesets labelled everyday: 1274
• Average changesets checked everyday: 37
• Average changesets harmful everyday: 4

[bkowshik]

The hypothesis behind the invalid-tag-modification compare function are:

• The primary tag of a feature is the identity of the feature.
• If a feature loose it's the primary tag, that means that it looses it's identity.
• A feature loosing it's identity could be problematic.

Good detections

Converting a residential area into a park

• http://osmcha.mapbox.com/47566781/
• http://osmcha.mapbox.com/47487736/
• landuse=residential -> leisure=park

Removal of uncommon tags.

• http://osmcha.mapbox.com/47617441/
• There are only 11,455 features on OpenStreetMap with amenity=grit_bin

False detections

Fix of a harmful edit gets flagged too

• http://osmcha.mapbox.com/47487736/ was problematic which was fixed in https://osmcha.mapbox.com/47497747/
• But, the fix was also flagged as there was a primary tag change as well, leisure=park -> landuse=residential

Tags moved from feature to relation

• http://osmcha.mapbox.com/47617390/
• As more features get mapped in greater detail, the more will be the number of relations.

Specialized value instead of specialized key

• http://osmcha.mapbox.com/47617324/
• Feature moves from amenity=college -> building=college
• There was duplicate values too in this case. amenity=college and building=college.
• http://osmcha.mapbox.com/47617351/
• Move from building: yes -> amenity: shelter

[maning]

False positive here: https://osmcha.mapbox.com/47616490/
My edits are the following which I believe are valid:

• removed hotel tag in a convenience store
• moved the landuse=residential name from node to polygon

[bkowshik]

I analyzed the 220 changesets checked on osmcha with 1,892 features to understand what primary tags are commonly deleted from features? Ex:

• The building primary tag was deleted from 879 features.
  

When primary tag of a feature is deleted:

• Is the feature given a new primary tag?
• If yes, which is the new primary tag?

For example, there were 8 occasions when landuse tag was deleted and building was added for the feature. Incidentally, 6 of the 8 were actually harmful. Ex: way!478636519 and relation!4656986.

2nd primary tag

Looking at quite a lot of modifications, I could infer that removing building=yes is a 👍 in scenarios when a feature has two or more primary tags, with one being building. Ex:

• way!479406193 with a second primary tag in amenity=social_facility
• way!485761848 with tourism=hotel
• way!259295888 with office=telecommunication

NOTE: Out of 879 building features flaged for invalid_tag_modification, 843 look like ^

Part of a relation

Feature's can be made part of a relation with all or some of it's properties removed.

• All properties removed: https://osmcha.mapbox.com/47442417/features/way-294035585/
• Some properties removed: http://osmcha.mapbox.com/47514902/features/way-4481547/

Feature changes like ^ mostly are good which we can 👍 for now. The problem is the feature does not have a reference to it being part of a relation. Any ideas to handle this scenario @geohacker @batpad @amishas157. The only hack I can think of is if all properties of a feature are removed, we could assume that the feature is becoming a part of a relation. This has been true for almost 100% of the modification's I have 👀.

# Example of a feature coming to the comparator.
# Note that there isn't any reference to the relation that it is part of.
{
    "type": "Feature",
    "id": "way!333194926!3",
    "properties": {
        "osm:type": "way",
        "osm:id": 333194926,
        "osm:version": 3,
        "osm:changeset": 47592706,
        "osm:timestamp": 1491745717000,
        "osm:uid": 665748,
        "osm:user": "sebastic"
    },
    "geometry": {
        "type": "LineString",
        "coordinates": [
            [
                -76.4374173,
                36.98812
            ],
            [
                -76.4372177,
                36.9877951
            ],
            [
                -76.4371429,
                36.9878247
            ],
            [
                -76.4370818,
                36.9877254
            ],
            [
                -76.4369996,
                36.9877579
            ],
            [
                -76.4372598,
                36.9881822
            ],
            [
                -76.4374173,
                36.98812
            ]
        ]
    }
}