parser = argparse.ArgumentParser(description="Run FM")
    parser.add_argument('-lr', type=float, metavar='<lr>', dest='lr', help='lr')
    parser.add_argument('-flr', type=float, metavar='<flr>', dest='flr', help='flr')
    # means the learning rate of feature similarity learning
    parser.add_argument('-reg', type=float, metavar='<reg>', dest='reg', help='reg')
    # regularization
    parser.add_argument('-decay', type=float, metavar='<decay>', dest='decay', help='decay')
    # weight decay
    parser.add_argument('-qonly', type=int, metavar='<qonly>', dest='qonly', help='qonly')
    # means quadratic form only (letting go other terms in FM equation...)
    parser.add_argument('-bs', type=int, metavar='<bs>', dest='bs', help='bs')
    # batch size
    parser.add_argument('-hs', type=int, metavar='<hs>', dest='hs', help='hs')
    # hidden size
    parser.add_argument('-ip', type=float, metavar='<ip>', dest='ip', help='ip')
    # init parameter for hidden
    parser.add_argument('-dr', type=float, metavar='<dr>', dest='dr', help='dr')
    # dropout ratio
    parser.add_argument('-optim', type=str, metavar='<optim>', dest='optim', help='optim')
    # optimizer
    parser.add_argument('-observe', type=int, metavar='<observe>', dest='observe', help='observe')
    # the frequency of doing evaluation
    parser.add_argument('-oldnew', type=str, metavar='<oldnew>', dest='oldnew', help='oldnew')
    # we don't use this parameter now
    parser.add_argument('-pretrain', type=int, metavar='<pretrain>', dest='pretrain', help='pretrain')
    # does it need to load pretrain model?
    parser.add_argument('-uf', type=int, metavar='<uf>', dest='uf', help='uf')
    # update feature
    parser.add_argument('-rd', type=int, metavar='<rd>', dest='rd', help='rd')
    # remove duplicate, we don;t use this parameter now
    parser.add_argument('-useremb', type=int, metavar='<useremb>', dest='useremb', help='user embedding')
    # update user embedding during feature similarity
    parser.add_argument('-freeze', type=int, metavar='<freeze>', dest='freeze', help='freeze')
    # we don't use this param now
    parser.add_argument('-command', type=int, metavar='<command>', dest='command', help='command')
    # command = 6: normal FM
    # command = 8: with our second type of negative sample
    parser.add_argument('-seed', type=int, metavar='<seed>', dest='seed', help='seed')
    # random seed
    A = parser.parse_args()

    parser = argparse.ArgumentParser(description="Run conversational recommendation.")
    parser.add_argument('-mt', type=int, dest='mt', help='MAX_TURN')
    parser.add_argument('-playby', type=str, dest='playby', help='playby')
    # options include:
    # AO: (Ask Only and recommend by probability)
    # RO: (Recommend Only)
    # policy: (action decided by our policy network)
    parser.add_argument('-fmCommand', type=str, dest='fmCommand', help='fmCommand')
    # the command used for FM, check out /EAR/lastfm/FM/
    parser.add_argument('-optim', type=str, dest='optim', help='optimizer')
    # the optimizer for policy network
    parser.add_argument('-lr', type=float, dest='lr', help='lr')
    # learning rate of policy network
    parser.add_argument('-decay', type=float, dest='decay', help='decay')
    # weight decay
    parser.add_argument('-TopKTaxo', type=int, dest='TopKTaxo', help='TopKTaxo')
    # how many 2-layer feature will represent a big feature. Only Yelp dataset use this param, lastFM have no effect.
    parser.add_argument('-gamma', type=float, dest='gamma', help='gamma')
    # gamma of training policy network
    parser.add_argument('-trick', type=int, dest='trick', help='trick')
    # whether use normalization in training policy network
    parser.add_argument('-startFrom', type=int, dest='startFrom', help='startFrom')
    # startFrom which user-item interaction pair
    parser.add_argument('-endAt', type=int, dest='endAt', help='endAt')
    # endAt which user-item interaction pair
    parser.add_argument('-strategy', type=str, dest='strategy', help='strategy')
    # strategy to choose question to ask, only have effect
    parser.add_argument('-eval', type=int, dest='eval', help='eval')
    # whether current run is for evaluation
    parser.add_argument('-mini', type=int, dest='mini', help='mini')
    # means `mini`-batch update the FM
    parser.add_argument('-alwaysupdate', type=int, dest='alwaysupdate', help='alwaysupdate')
    # means always mini-batch update the FM, alternative is that only do the update for 1 time in a session.
    # we leave this exploration tof follower of our work.
    parser.add_argument('-initeval', type=int, dest='initeval', help='initeval')
    # whether do the evaluation for the `init`ial version of policy network (directly after pre-train)
    parser.add_argument('-upoptim', type=str, dest='upoptim', help='upoptim')
    # optimizer for reflection stafe
    parser.add_argument('-upcount', type=int, dest='upcount', help='upcount')
    # how many times to do reflection
    parser.add_argument('-upreg', type=float, dest='upreg', help='upreg')
    # regularization term in
    parser.add_argument('-code', type=str, dest='code', help='code')
    # We use it to give each run a unique identifier.
    parser.add_argument('-purpose', type=str, dest='purpose', help='purpose')
    # options: pretrain, others
    parser.add_argument('-mod', type=str, dest='mod', help='mod')
    # options: CRM, EAR
    parser.add_argument('-mask', type=int, dest='mask', help='mask')
    # use for ablation study, 1, 2, 3, 4 represent our four segments, {ent, sim, his, len}