We prepared a help file which is part of the ROCA toolbox and this FAQ page with Frequently Asked Questions. If you have further questions, contact us by e-mail.

How to prepare your own training dataset

The default training dataset consists of several secondary roads from the Czech road network. It is prepared from our road network data, so it is primarily suitable for data having similar quality as our Czech road network.
 
Your road network should be divided into road sections by intersections. Each road section can be composed of several curves and tangents.

It is convenient to prepare your own training dataset from your own road network while keeping the same structure as the default training dataset. It should be a small part of your road network (polyline feature class - shapefile). We prepared training dataset from 32 road section geometries (together 2385 vertices), which is approx 0,1 % of the entire Czech road network.

You can use the following process:

1) select (randomly) few polyline features from your network shapefile,
2) convert these polyline features to vertices (tool Feature Vertices to Points in Data Management of ArcGIS),
3) than for each vertex in this list, specify if it is a part of tangent geometry (0) or horizontal curve (1) - based on your data quality and/or your expert opinion,
4) save as your own training .txt file (with any consecutive whitespaces - tab, space - as delimiter of columns).

You have to include points (vertices of polylines) of both classes (part of horizontal curves and part of tangents). The ratio between points with class 1 and class 0 in your txt. training file is recommended to be 1:1 (or at least in a range 1:4 to 4:1).

You can use the default Czech training dataset in case that your network have similar quality as the Czech dataset.

Additional questions

Does the sample of vertices in the training dataset correspond to one or more features in the 'Input line feature class' dataset?
Yes

For the training file, do I need to create vertices from polyline road section that I choose and decide whether every vertex belongs to a tangent or a curve (0/1)?
Yes
 
How to specify vertices which are part of a reverse curve (two opposite curves) or a compound curve for training dataset?

If you have a reverse or compound curve on your road segment, there is no need to put 0 between them. Specify all vertices of reverse curve as 1. In general, there is no need to put 0 in between the two curves. The algorithm will decide if there is a compound curve or a reverse curve on the basis of radius of osculating circle drawn at each vertex (having the same tangent and normal as the road).

Where can I get the detour ratio, number of turns along a road section, and the average cumulative angle turned per km?

This new three fields with curvature attributes will be added to the original input line feature class.