Reorganization of abc related algorithms. ABC SMC with observations is implemented but not yet tested. automaton abc still works.

The two Cosmos tests checks the statistical correctness of estimated
distance value for R1,R2,R3,R5 experiments. Also, they test that all
trajectories are accepted.

A bunch of errors were fixed in automata files (Cosmos .lha and julia
./automata files) and simulation function that introduced statistical
bias, i.e. approximated values that were closed but not in the
confidence interval. The general structure of simulation and was
reworked in order to suits well te behavior of Cosmos.

copyto! is implemented for LHA states.

Automaton G works almost well, a bias in the distance is identified
compared to Cosmos when one of the model parameters is zero.

Implementation of volatile_simulate, which simulates SynchronizedModel
without saving values: it only returns the last LHA state of the
simulation.

Implementation of distributed computations of variable automata with
volatile_simulate.

Tests passed.

R1,R2,R3.

Fix about methods of SynchronizedTrajectory + tests.

Simulation for ParametricModel works.

efficient.

First bricks of the automaton_abc algorithm.

slight improvement in allocation and time in benchmark but too small.

small.

Fix of a misbehaviour of isbounded() for trajectories.

Add several convenient methods for parameters of model.

Next step: implementation of simulate with resize!.

order to replace vcat by push!.

The benchmarks are at least equal, and better when the buffer size is
small. However, no significative difference of performance with adequate
buffer size.

My opinion the gain can be seen when simulations are going to be very
long with more complicated models than we have implemented for now.

All tests passed.

Automaton G is implemented and runs OK. Not tested with Cosmos call.

The next step of development is the restruction of data collection for
trajectories.

From now on: Trajectory.values::Matrix{Float64}, it will be
Trajectory.values::Vector{Vector{Float64}}.

According to small performance tests I've made, it will increase the
performance of the package.

Add of set_time_bound!
Add of a first notebook example to explain the package.

Add of a major test based on Cosmos: tests/cosmos/distance_F/ER_1D.jl

The LHA distance F of this package is tested with Cosmos.
For several parameters in R1, R2, R3 experiments (see Bentriou,
Ballarini, Cournede 2018), Cosmos is called to
compute an average distance. Then we collect the number of simulations
and the estimated value computed by Cosmos.
We compute an average distance by the package methods with the same
number of simulations done by Cosmos

Then the two average distances are compared to each others with
regard to the width of the confidence interval set in Cosmos.
Test passed.

Add of a new type of Trajectory: SynchronizedTrajectory which is created
by the simulation of a synchronized model with an LHA.

Simulate function for SynchronizedModel executes and is tested by two
simple models.

Next step : Cosmos based tests.

Add of several useful functions for access values of model, lha and
trajectory objects.

an object.

Type stability of simulate, dist, read_trajectory seems to be preserved.

First implementation of LHA.
Synchronized simulation not yet tested.
Reading of trajectories with LHA seems to work with one test basted on
simulations that checks that the last value is well read.

Major fix of type unstabilities which improved performances.
For now on, we observe for pygmalion benchmarks:
+2x speed simulation compared to simulation of models.
100x speed in terms of reading values (no allocations).

Add of bench for @view macro.

Fix of a major issue in the simulation: sometimes the number of states
wasn't consistent. Add of test to check the consistency of the
Trajectory structure (unit/check_trajectory_consistency).

Implementation of the lp distance for trajectories.
Add of unit tests for lp distance.

can be set by the user. Changes of SIR and ER.

previously loaded model). The package warns when this happen.

Several benchmarks for the data structure. Now we chose the row buffer
method according to results (mainly because the trajectory access with
far way faster, but strangly not that much differences in other cases).