# CSR Experiments

- Vorschlag von Thomas:
- feasibility von Pareto Dijsktra (Laufzeit)
- warum feasible (Pareto-Front Daten)
- Vergleich gegen selfish routing
- Mehrere Datensaetze
- selected OD-pairs
- common sections

## Data sets

| Name | Description | Variables |
| -------- | -------- | -------- |
| **kc-sections** | shared sections of shortest paths of OD-pairs | $k$, timeframe, base data |
| **single-pairs** | multiplied OD-pairs | $d$ (number of drivers) |

## Experiment strategy

- Measurements
- Runtime (cpu-time)
- Quality ()
- Pareto-Front size

- **kc-sections**
- base data:
- approx. user equilibrium dijkstra
- $k$: 100 - 500, step_size = 10
- $k$-common strings longer than 50 nodes

## Scenario list

| dataset | $k$ (#drivers) | origin node | destination node |
| - | - | - | - |

For each row a plans.xml can be generated.

## Result
csv

dataset $\in$ (kcs(9|12|17)|ttp(9|12|17))
strategy $\in$ (1D-SAP, UD-SAP, ShortestPath)

| dataset | $k$ (#drivers) | origin node | destination node | actual $k$ | psychmod | strategy | cpu-time | score | # node visits | # canditdates | # edges in original route | mean Pareto set size | sum Pareto set size | Mean DP size | Sum DP size | Found alternative (true=0/false=1)
| - | - | - | - | - | - | - | - | - | - | - | - | - | - | - | - | - |

DP size references to the size of the fields of the Array A in the DP. If the strategy is not nodisjoint, these values are 0.

Unclear: Also test strategy = CD-SAP? Problem: Most of the OD-pairs won't work

## File structure
input:
- in.csv
- sane_berlin.xml
- kcs
- origin_destination_k.xml
- origin_destination_k.xml
- origin_destination_k.xml
- ttp
- origin_destination_k.xml
- origin_destination_k.xml
- origin_destination_k.xml

output:
- kcs
- origin_destination_k_routed_strategy_psych.xml
- ttp
- origin_destination_k_routed_strategy_psych.xml
- stats.csv

## Paper Plots

- Runtime (grouped bar plots)
- by Dataset (color)
- by Strategy (x-dimension)

- Runtime for $k$ (scatter plot)
- by $k$ (x-dimension)
- by Strategy (color)
- by Dataset (two plots)

- Quality (grouped bar plots)
- by Dataset (color)
- by Strategy (also Dijkstra-1 and Dijkstra-All) (x-dimension)

- Quaility boost by $k$ -> relative improvement to DijkstraAll (histogram) (if interesting)
- by Dataset (two plots)
- by Strategy (CD, 1D, UD)

- Density on RouteLength vs. $k$ plane (2D-Histogram)

Psychologisches Modell? -> ggf. auf user_equilibrium beschraenken

## Hinweise zum Text

* Tomtom skalierter Graph in den Fliesstext mit aufnehmen, damit das Setting klar ist
* Bogen schlagen zu Section 2
* Hardwarebeschreibung / Implementierungsbeschreibung
* textueller Hinweis auf die andere psychologischen Modelle

## Feedback Louise + Thomase
### Plots
- Score-Kurven der Strategien über k
- Runtimes der Strategien über Distanz (Buckets oder einzelne Punkte)

- Runtime über Pareto-Fronten, wenn ähnlich zu anderem Runtime

- Plots jeweils nach Psychmod

### Text
in den Draft soll auch das intelligente Insert and Dominate, Multi-criteria A*,
schon Detailliert (aber schnell beschreiben)

Implementationdetails + Evaluation