Esempio n. 1
0
def make_exp(env_pairs,
             cfg=None,
             nor_diff={}, src_diff={}, tgt_diff={},
             path='phd/dov', prefix=('dov',),
             nor_ex_names=(),
             src_ex_names=(),
             tgt_ex_names=()):

    nor_diff = copy.deepcopy(nor_diff)
    src_diff = copy.deepcopy(src_diff)
    tgt_diff = copy.deepcopy(tgt_diff)


    trio_exps = []

    for src_env_name, tgt_env_name in env_pairs:

        # no reuse experiment confs
        nor_exps = []
        if len(nor_ex_names) == 0:
            nor_exps.append(None)
        # else
        for nor_ex_name in nor_ex_names:
            nor_cfg_i = cfg._deepcopy()
            nor_cfg_i.exploration.explorer._update(exs.catalog[nor_ex_name], overwrite=False)
            nor_cfg_i.exploration.env._update(envs.catalog[tgt_env_name], overwrite=False)
            nor_cfg_i.exploration.seeds = factored.src_seeds
            nor_cfg_i._update(nor_diff, overwrite=True)

            nor_cfg_i.exploration.ex_name  = nor_ex_name
            nor_cfg_i.exploration.env_name = tgt_env_name

            nor_exps.append(nor_cfg_i)


        # source experiment confs
        norsrc_exps = []
        for nor_cfg_i in nor_exps:
            if len(src_ex_names) == 0:
                norsrc_exps.append((nor_cfg_i, None))
            # else
            for src_ex_name in src_ex_names:
                src_cfg_i = cfg._deepcopy()
                src_cfg_i.exploration.explorer._update(exs.catalog[src_ex_name], overwrite=False)
                src_cfg_i.exploration.env._update(envs.catalog[src_env_name], overwrite=False)
                src_cfg_i.exploration.seeds = factored.src_seeds
                src_cfg_i._update(src_diff, overwrite=True)

                src_cfg_i.exploration.ex_name  = src_ex_name
                src_cfg_i.exploration.env_name = src_env_name

                norsrc_exps.append((nor_cfg_i, src_cfg_i))


        # target experiment confs
        for nor_cfg_i, src_cfg_i in norsrc_exps:
            if len(tgt_ex_names) == 0:
                trio_exps.append((nor_cfg_i, src_cfg_i, None))
            # else
            for tgt_ex_name in tgt_ex_names:
                tgt_cfg_i = cfg._deepcopy()
                tgt_cfg_i.exploration.explorer._update(exs.catalog[tgt_ex_name], overwrite=False)
                tgt_cfg_i.exploration.env._update(envs.catalog[tgt_env_name], overwrite=False)
                tgt_cfg_i.exploration.seeds = factored.tgt_seeds

                tgt_cfg_i.exploration.deps = (experiments.expkey(src_cfg_i),)
                tgt_cfg_i._update(tgt_diff, overwrite=True)
                tgt_cfg_i.exp.prefix = tuple(tgt_cfg_i.exp.prefix) + (src_cfg_i.exploration.ex_name, src_cfg_i.exploration.env_name)

                tgt_cfg_i.exploration.ex_name  = tgt_ex_name
                tgt_cfg_i.exploration.env_name = tgt_env_name

                trio_exps.append((nor_cfg_i, src_cfg_i, tgt_cfg_i))


    return trio_exps
Esempio n. 2
0
    r_ex          = explorers.MetaExplorer.defcfg._deepcopy()
    r_ex.eras     = (mb, None)
    r_ex.weights  = ((1-p_reuse, 0.0, p_reuse), (0.0, 1.0, 0.0))

    r_ex.ex_0                   = explorers.RandomMotorExplorer.defcfg._deepcopy()

    r_ex.ex_1                   = explorers.RandomGoalExplorer.defcfg._deepcopy()
    r_ex.ex_1.learner           = learners.MutateNNLearner.defcfg._deepcopy()
    r_ex.ex_1.learner.operator.name = 'uniform'
    r_ex.ex_1.learner.operator.d    = 0.05

    r_ex.ex_2                   = explorers.ReuseExplorer.defcfg._deepcopy()
    r_ex.ex_2.reuse.res         = res
    r_ex.ex_2.reuse.algorithm   = algorithm

    alg_str = '' if algorithm == 'sensor_uniform' else '.' + algorithm
    return 'reuse{}_{}_{}_{}_{}'.format(alg_str, mb, p_reuse, res, lrn_name), r_ex


tgt_cfg = cfg._deepcopy()
tgt_cfg.exp.path = 'unit_tests/'

tgt_cfg.exploration.ex_name, tgt_cfg.exploration.explorer = reuse_ex(50, 0.5)
tgt_cfg.exploration.deps = (experiments.expkey(cfg),)
tgt_cfg.exp.prefix   = ('prefix2',) #, 'random.motor', 'kin7_150')


if __name__ == '__main__':
    experiments.run_exps([cfg, tgt_cfg])
Esempio n. 3
0
    r_ex = explorers.MetaExplorer.defcfg._deepcopy()
    r_ex.eras = (mb, None)
    r_ex.weights = ((1 - p_reuse, 0.0, p_reuse), (0.0, 1.0, 0.0))

    r_ex.ex_0 = explorers.RandomMotorExplorer.defcfg._deepcopy()

    r_ex.ex_1 = explorers.RandomGoalExplorer.defcfg._deepcopy()
    r_ex.ex_1.learner = learners.MutateNNLearner.defcfg._deepcopy()
    r_ex.ex_1.learner.operator.name = 'uniform'
    r_ex.ex_1.learner.operator.d = 0.05

    r_ex.ex_2 = explorers.ReuseExplorer.defcfg._deepcopy()
    r_ex.ex_2.reuse.res = res
    r_ex.ex_2.reuse.algorithm = algorithm

    alg_str = '' if algorithm == 'sensor_uniform' else '.' + algorithm
    return 'reuse{}_{}_{}_{}_{}'.format(alg_str, mb, p_reuse, res,
                                        lrn_name), r_ex


tgt_cfg = cfg._deepcopy()
tgt_cfg.exp.path = 'unit_tests/'

tgt_cfg.exploration.ex_name, tgt_cfg.exploration.explorer = reuse_ex(50, 0.5)
tgt_cfg.exploration.deps = (experiments.expkey(cfg), )
tgt_cfg.exp.prefix = ('prefix2', )  #, 'random.motor', 'kin7_150')

if __name__ == '__main__':
    experiments.run_exps([cfg, tgt_cfg])