196
X-Machines for Agent-Based Modeling: FLAME Perspectives
fputs("\n", file);
(void)fclose(file);
return 0;
}
7.3.3 Find Optimum Model Settings
To ensure a correct comparison of the two techniques, both models were
simulated with identical conditions and data were collected and analyzed. The
experience in both simulation techniques was compared across a number of
factors like simulation time, the memory size needed and tools used (Table
7.1).
Analyzing the time to actually write the models can be arguable, depending on the experience of the programmers. A programmer with little prior
knowledge of agent-based modeling may take more than a month to get accustomed to agents over the platforms. This also includes an installation and
learning time for the actual agent-based platforms. For stochastic simulation,
MATLAB specializes in mathematical function writing, and would be relatively easier to grasp than a different agent-based modeling framework.
Both models produce results in different ways. Agent-based models produce results as time snapshots for agent conditions at these times. MATLAB
can produce concentration gradients, to see the overall system behavior at
different times.
The global values can be another deciding factor in how the results look
in the end. These need to be tested with multiple runs, to find optimum
conditions for the simulation to give results which match closest to real data.
Agent-based model results can also use averages over a number of simulations
runs, to compensate for the random nature of the agents inherent in the
models.
Figure 7.7 shows the intensity plots of the protein distribution across the
embryo during the simulation. The figure shows plots (from left to right)
deterministic, stochastic and agent-based models. Another representation is
shown in Figure 7.8, where peaks of molecule numbers in compartments are
compared.
The results show the decay rate being too high in the agent model, where
the protein agents die before reaching the last end of the embryo cell (last compartment). Therefore this needs to be reduced in order to match the stochastic behavior. Therefore, using the same initial conditions from the stochastic
model, the results are not able to be replicated in the agent-based simulations.
Figure 7.9 shows the missing data points in the resulting figures when both
models use the same initial setting. The agent-based model is then simulated
multiple times with different global conditions to find the best set of values
that will produce results closer to the stochastic model (Table 7.2).
X-Machines for Agent-Based Modeling: FLAME Perspectives
fputs("\n", file);
(void)fclose(file);
return 0;
}
7.3.3 Find Optimum Model Settings
To ensure a correct comparison of the two techniques, both models were
simulated with identical conditions and data were collected and analyzed. The
experience in both simulation techniques was compared across a number of
factors like simulation time, the memory size needed and tools used (Table
7.1).
Analyzing the time to actually write the models can be arguable, depending on the experience of the programmers. A programmer with little prior
knowledge of agent-based modeling may take more than a month to get accustomed to agents over the platforms. This also includes an installation and
learning time for the actual agent-based platforms. For stochastic simulation,
MATLAB specializes in mathematical function writing, and would be relatively easier to grasp than a different agent-based modeling framework.
Both models produce results in different ways. Agent-based models produce results as time snapshots for agent conditions at these times. MATLAB
can produce concentration gradients, to see the overall system behavior at
different times.
The global values can be another deciding factor in how the results look
in the end. These need to be tested with multiple runs, to find optimum
conditions for the simulation to give results which match closest to real data.
Agent-based model results can also use averages over a number of simulations
runs, to compensate for the random nature of the agents inherent in the
models.
Figure 7.7 shows the intensity plots of the protein distribution across the
embryo during the simulation. The figure shows plots (from left to right)
deterministic, stochastic and agent-based models. Another representation is
shown in Figure 7.8, where peaks of molecule numbers in compartments are
compared.
The results show the decay rate being too high in the agent model, where
the protein agents die before reaching the last end of the embryo cell (last compartment). Therefore this needs to be reduced in order to match the stochastic behavior. Therefore, using the same initial conditions from the stochastic
model, the results are not able to be replicated in the agent-based simulations.
Figure 7.9 shows the missing data points in the resulting figures when both
models use the same initial setting. The agent-based model is then simulated
multiple times with different global conditions to find the best set of values
that will produce results closer to the stochastic model (Table 7.2).