Skip to content

Latest commit

 

History

History
21 lines (13 loc) · 1.67 KB

README.md

File metadata and controls

21 lines (13 loc) · 1.67 KB
Raw

LaCIM

  1. Simulation

To get the results of the second row in Table 1, 1). run lacim_d_num_cluster_3_original_1667.py 2). np.mean(z_perf_ivae_discrete_3,axis=1), np.mean(s_perf_ivae_discrete_3,axis=1)

To get the results of the third row in Table 1,

  1. run lacim_d_num_cluster_5.py
  2. np.mean(z_perf_ivae_discrete_5,axis=1), np.mean(s_perf_ivae_discrete_5,axis=1)
  1. Real World (Colored MNIST)

  • python generate_colored_mnist.py --sigma 0.02 --env_num 2 --color_num 2 --env_type 0 --test_ratio 0.1 to generate colored MNIST. The training environment include 0.8 and 0.9. The ratio of testing environment is 0.1.

  • python generate_colored_mnist.py --sigma 0.02 --env_num 2 --color_num 2 --env_type 3 --test_ratio 0.1 to generate colored MNIST. The training environment include 0.9 and 0.95. The ratio of testing environment is 0.1.

  • CUDA_VISIBLE_DEVICES=0 python LaCIM_rho.py --epochs 120 --optimizer sgd --lr 0.1 --lr_decay 0.5 --lr_controler 80 --in_channel 3 --batch-size 256 --test-batch-size 256 --reg 0.0002 --beta 1 --dataset mnist --num_classes 2 --env_num 2 --seed -1 --zs_dim 32 --root ./data/colored_MNIST_0.02_env_2_0_c_2_0.10/ --test_ep 50 --lr2 0.0005 --reg2 0.005 --sample_num 10 --image_size 28 --z_ratio 0.5 to get the results of ours in Table 3.

  • CUDA_VISIBLE_DEVICES=0 python d_LaCIM.py --epochs 200 --optimizer sgd --lr 0.3 --lr_decay 0.5 --lr_controler 120 --in_channel 3 --batch-size 256 --test-batch-size 256 --reg 0.0005 --dataset mnist --num_classes 2 --env_num 2 --seed -1 --zs_dim 32 --root ./data/colored_MNIST_0.02_env_2_3_c_2_0.10/ --test_ep 100 --lr2 0.007 --reg2 0.08 --sample_num 10 --image_size 28 --alpha 8.0 --gamma 1.0 --beta 1.0 --z_ratio 0.5