state.lr = 8.2 * 1e-3 state.batch_size = 160 state.sub_mb_size = 160 state.max_iters = 90000 state.n_hids = 180 state.mem_nel = 120 state.mem_size = 28 state.renormalization_scale = 5.0 state.use_ff_controller = True state.seed = args.seed state.use_gru_inp = True state.use_bow_inp = False state.std = 0.034 state.bow_size = 100 state.bow_weight_start = 0.64 state.learn_h0 = False state.n_reading_steps = 1 state.n_read_heads = 1 state.max_seq_len = 300 state.max_fact_len = 15 state.use_reinforce_baseline = False state.use_reinforce = False state.address_size = 16 state.path = "/rap/jvb-000-aa/data/sarath/en-10k/splitted_trainval/" state.lambda1_rein = 9e-5 state.lambda2_rein = 1e-5
state = DD() state.lr = 3e-3 state.batch_size = 160 state.sub_mb_size = 160 state.std = 0.05 state.max_iters = 40000 state.n_hids = 240 state.mem_nel = 150 state.mem_size = 28 state.renormalization_scale = 5.0 state.bowout = True state.use_ff_controller = False state.std = 0.01 state.bow_size = 80 state.n_reading_steps = 1 state.n_read_heads = 1 state.max_seq_len = 300 state.max_fact_len = 15 state.use_reinforce_baseline = False state.use_reinforce = True state.debug = False state.address_size = 24 state.path = "/scratch/jvb-000-aa/gulcehre/tasks_1-20_v1-2/en-10k/" state.lambda1_rein = 3e-5 state.lambda2_rein = 1e-5 state.theano_function_mode = None #NanGuardMode(nan_is_error=True, inf_is_error=True) state.print_every = 50 np.random.seed(3)
