Parameters: Evolution mode
The original's directed-evolution modes (hidden predator, ZeroBot, test) and the weight the app uses to pick the fittest bot.
The original DarwinBots had an evolution mode: an automatic procedure for evolving a bot from another one, which started simulations, saved files and restarted itself for hours on end. Of all that, the app keeps what happens inside the simulation, which is what the first three parameters in this group control. The outer procedure (the files, the stages, the automatic restart) isn't there: when one of these simulations reaches its end, it stops and that's it.
These are modes to experiment with carefully: they depend on species with fixed names and on values the app leaves at 0. The fourth parameter, on the other hand, is for everyday use: it decides which bot Find the best picks in Observe.
The modes
Restart mode picks the mode. The values with an effect inside the simulation are these; the others behave like 0.
| Value | Mode | What it does in the simulation |
|---|---|---|
| 0 | Normal | Nothing: an ordinary simulation. |
| 1 | Seeding | The disqualification rules apply as in an F1 contest. |
| 4 and 5 | Hidden predator | Alternates epochs in which the Base species disappears from the world and the Mutate species gets an energy boost (see below). The simulation stops if either of the two goes extinct. |
| 7 and 8 | ZeroBot | Every 50 cycles it looks for the fittest bot; if it belongs to the Mutate species, has already mutated and is a different bot from last time or has improved its score, it raises the mutation rates. It stops if Mutate goes extinct, or if the best is still the same bot and has improved its score (in the original, that is where it moved on to the test stage). |
| 9 | Test | Compares the energy of the Test species at cycle 1 and at cycle 8000: if it doubled and there are more than 10 animals, the test passes. At cycle 8000 it stops. |
Species are recognized by their exact name: they have to be called Base, Mutate or Test in the scenario.
The hidden predator
The idea of mode 4 is to train one species, Mutate, against a fixed rival, Base, without the rival wiping it out while it learns. To do that, the simulation alternates two epochs:
- With the predator in view. The two species coexist and compete normally.
- With the predator hidden. Base's bots are frozen and out of the world: they don't run their DNA, they don't move, nobody sees them and they can't be shot. Meanwhile, Mutate's bots receive an energy compensation every cycle, in full for those that have just mutated and half for the rest.
When a hidden epoch ends, before Base comes back, the feeding shots in flight are cleared and any Mutate bots that ended up too close to a Base bot are moved away, so the new epoch doesn't start with a point-blank attack.
The mode ends badly if Mutate goes extinct and well if Base does. In both cases the simulation stops; in the original, on winning, the fittest bot was saved (using the criterion of Offspring weight), and in the app that step isn't there.
Restart mode
Evolution loop mode (0 = normal; 4/5 = hidden predator).
- Default
- 0
- Usual range
- 0 to 9
- Variable
x_restartmode- Key
opt:92- Live
- can be changed while the simulation runs
- Mode
- advanced only
Picks the directed-evolution mode from the table above. By default it is 0, a normal simulation, and that is what you want unless you are reproducing an experiment from the original. Modes 4 and 5 do the same thing inside the simulation, as do 7 and 8: in the original they were told apart by what the outer procedure did. If the species aren't named the way the mode expects, mode 4 stops in the first cycle because it can't find Mutate.
Hidden predator cycles
Length of each hidden-predator epoch (between 0.83 and 1.17 times this value) and of its energy ramp. At 0 the mode changes every cycle.
- Default
- 0
- Usual range
- 0 to 32000
- Variable
hidePredCycl- Key
opt:94- Live
- can be changed while the simulation runs
- Mode
- advanced only
How long each hidden-predator epoch lasts, in cycles: each one lasts between 0.83 and 1.17 times this value, with a random part so the bots can't learn the schedule. It also sets the ramp of the energy boost, which grows from 0 to its full value during the first (8 × this value) cycles of the simulation. With 0 (the factory value), the epochs change every cycle and the mode is useless; to use it, set a few thousand.
Energy factor (LFOR)
Divisor of the hidden predator energy.
- Default
- 0
- Usual range
- 0 to 150
- Variable
LFOR- Key
opt:95- Live
- can be changed while the simulation runs
- Mode
- advanced only
The divisor of Mutate's energy boost. The engine compares how much the energy of newly mutated bots changes per cycle in one epoch and in the other, and while the predator is hidden it gives them that difference divided by this number: with a larger value, the boost is smaller and evolution is more demanding. With 0 (the factory value) the boost isn't calculated and Mutate gets nothing. At the cap, 150, the hidden epoch stretches out for as long as there are fewer Mutate bots than Base bots.
Offspring weight
To pick the best bot: 100 = balance between the bot and its offspring.
- Default
- 100
- Usual range
- 0 to 200
- Variable
intFindBestV2- Key
opt:96- Live
- can be changed while the simulation runs
- Mode
- advanced only
How the fittest bot is picked. For each bot that is neither a vegetable nor a corpse, the engine adds up its invested energy (energy plus 10 per point of body) and that of all its living descendants, up to ten generations, and combines that total with the number of descendants:
- at 100 (the factory value), both count equally: the score is the family's invested energy multiplied by (the number of descendants + 1);
- below 100, energy weighs less, and at 0 only the number of descendants counts;
- above 100, descendants weigh less, and at 200 only the family's energy counts.
This value decides which bot Find the best selects in Observe (Observe) and the Fitness column of the snapshot of the living and of the record of the dead (Parameters: Recording). In ZeroBot and hidden-predator modes it is also the criterion the engine uses to pick the best.