def run2():
rankers = []
ranker_params = {
'learning_rate': 0.1,
'learning_rate_decay': 0.9999977,
}
sim_args, other_args = parser.parse_all_args(ranker_params)
run_name = 'PDGD'
rankers.append((run_name, PDGD, other_args))
run_name = 'DeepPDGD'
rankers.append((run_name, DeepPDGD, other_args))
sim = DataSimulation(sim_args)
sim.run(rankers)
def run1():
rankers = []
# NSGD with document space projection
ranker_params = {
'learning_rate_decay': 0.9999977,
'k_initial': 3,
'k_increase': False,
'GRAD_SIZE':60,
'EXP_SIZE':25,
'TB_QUEUE_SIZE':10,
'TB_WINDOW_SIZE':50,
'prev_qeury_len': 10}
sim_args, other_args = parser.parse_all_args(ranker_params)
run_name = 'DSGD_TD_NSGD'
rankers.append((run_name, TD_NSGD_DSP, other_args))
sim = DataSimulation(sim_args)
sim.run(rankers)
def func_pairrank(args, dir_name):
ranker_params = {
"learning_rate": args.lr,
"learning_rate_decay": args.lr_decay,
"update": args.update,
"_lambda": args.lmbd,
"alpha": args.alpha,
"refine": args.refine,
"rank": args.rank,
"ind": args.ind,
}
sim_args, other_args = parser.parse_all_args(ranker_params)
if args.update == "gd" or args.update == "gd_diag" or args.update == "gd_recent":
ranker_name = "None-None-{}-{}-{}-{}-{}-{}-{}".format(
args.update,
args.lmbd,
args.alpha,
args.refine,
args.rank,
args.ind,
args.seed,
)
else:
ranker_name = "{}-{}-{}-{}-{}-{}-{}-{}-{}".format(
args.lr,
args.lr_decay,
args.update,
args.lmbd,
args.alpha,
args.refine,
args.rank,
args.ind,
args.seed,
)
run_name = dir_name + ranker_name
ranker = [(run_name, PairRank, other_args)]
sim = DataSimulation(sim_args)
sim.run(ranker)
rankers = []
for vec in [50]:
parser.set_argument('n_embedding_features', vec)
ranker_params = {'conv_hist': 10}
arg_str, args, sim_args, mgd_args, emb_args = parser.parse_all_args(ranker_params)
run_name = 'PMGD19cand'
rankers.append((arg_str, run_name, ProbMeanBandit, [mgd_args], {}))
# run_name = 'DocSim_StaticKMeans_%dvectors' % vec
# rankers.append((arg_str, run_name, StaticKMeans, [emb_args, mgd_args], ranker_params))
#
# ranker_params = {'gradient_weight': float(0), 'kernel': 'linear'}
# arg_str, args, sim_args, mgd_args, emb_args = parser.parse_all_args(ranker_params)
# run_name = 'DocSim_linear_%dvectors' % vec
# rankers.append((arg_str, run_name, NormalizedSVMBandit, [emb_args, mgd_args], ranker_params))
#
# ranker_params = {'conv_hist': 10, 'change_threshold': 0.01, 'linear_renorm': False}
# arg_str, args, sim_args, mgd_args, emb_args = parser.parse_all_args(ranker_params)
# run_name = 'DocSim_cascade%dhist%sthres_linear_%dvectors' % (
# ranker_params['conv_hist'], ranker_params['change_threshold'], vec)
# rankers.append((arg_str, run_name, DocSim, [emb_args, mgd_args], ranker_params))
#
# ranker_params = {'conv_hist': 10, 'change_threshold': 0.01, 'linear_renorm': False}
# arg_str, args, sim_args, mgd_args, emb_args = parser.parse_all_args(ranker_params)
# run_name = 'DocSim_cascadeKMeans_%dvectors' % (vec)
# rankers.append((arg_str, run_name, CascadeKMeans, [emb_args, mgd_args], ranker_params))
sim = DataSimulation(sim_args)
sim.run(rankers)