-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathtraining_commands.json
19 lines (15 loc) · 12.9 KB
/
training_commands.json
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
{
"cose" : "python ink_training_eager_predictive.py --experiment_id 1590486000.6 --comment 'tf2-af03-t_s4-hyb32-tres2-gt_targ-fullstopg' --gt_targets --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 4 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model transformer --transformer_scale --transformer_pos_encoding --transformer_layers 6 --transformer_heads 4 --transformer_dmodel 64 --transformer_hidden_units 256 --transformer_dropout 0.0 --latent_units 8 --decoder_model t_emb --decoder_dropout 0.0 --decoder_layers 4 --decoder_hidden_units 512,512,512,512 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position --disable_pen_loss --mask_encoder_pen",
"cose-32" : "python ink_training_eager_predictive.py --experiment_id 1590706693.8 --comment 'tf2-af03-t_s4-hyb32-tres2-gt_targ-fullstopg' --gt_targets --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 4 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model transformer --transformer_scale --transformer_pos_encoding --transformer_layers 6 --transformer_heads 4 --transformer_dmodel 64 --transformer_hidden_units 256 --transformer_dropout 0.0 --latent_units 32 --decoder_model t_emb --decoder_dropout 0.0 --decoder_layers 4 --decoder_hidden_units 512,512,512,512 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position --disable_pen_loss --mask_encoder_pen",
"cose-mse" : "python ink_training_eager_predictive.py --experiment_id 1591122336.4 --comment 'tf2-af03-t_s4-hyb32-tres2-gt_targ-fullstopg' --gt_targets --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss mse --n_t_samples 4 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model transformer --transformer_scale --transformer_pos_encoding --transformer_layers 6 --transformer_heads 4 --transformer_dmodel 64 --transformer_hidden_units 256 --transformer_dropout 0 --latent_units 8 --decoder_model t_emb --decoder_dropout 0.0 --decoder_layers 4 --decoder_hidden_units 512,512,512,512 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position --disable_pen_loss --mask_encoder_pen",
"cose-mse-t100" : "python ink_training_eager_predictive.py --experiment_id 1591122336.6 --comment 'tf2-af03-100-hyb32-tres2-gt_targ-fullstopg' --gt_targets --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss mse --n_t_samples 100 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model transformer --transformer_scale --transformer_pos_encoding --transformer_layers 6 --transformer_heads 4 --transformer_dmodel 64 --transformer_hidden_units 256 --transformer_dropout 0 --latent_units 8 --decoder_model t_emb --decoder_dropout 0.0 --decoder_layers 4 --decoder_hidden_units 512,512,512,512 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position --disable_pen_loss --mask_encoder_pen",
"cose-cats" : "python ink_training_eager_predictive.py --experiment_id 1590979999.1 --comment 'tf2-af02-t_s4-hyb32-tres2-gt_targ-fullstopg' --gt_targets --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 4 --batch_size 128 --affine_prob 0.2 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model transformer --transformer_scale --transformer_pos_encoding --transformer_layers 6 --transformer_heads 4 --transformer_dmodel 64 --transformer_hidden_units 256 --transformer_dropout 0.0 --latent_units 8 --decoder_model t_emb --decoder_dropout 0.0 --decoder_layers 4 --decoder_hidden_units 512,512,512,512 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name quickdraw_cats --metadata_type position --disable_pen_loss --mask_encoder_pen",
"cose-elephant" : "python ink_training_eager_predictive.py --experiment_id 1591260847.1 --comment 'tf2-af02-t_s4-hyb32-tres2-gt_targ-fullstopg' --gt_targets --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 4 --batch_size 128 --affine_prob 0.2 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model transformer --transformer_scale --transformer_pos_encoding --transformer_layers 6 --transformer_heads 4 --transformer_dmodel 64 --transformer_hidden_units 256 --transformer_dropout 0.0 --latent_units 8 --decoder_model t_emb --decoder_dropout 0.0 --decoder_layers 4 --decoder_hidden_units 512,512,512,512 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name quickdraw_elephant --metadata_type position --disable_pen_loss --mask_encoder_pen",
"cose_enc-vae+cose_dec" : "python ink_training_eager_predictive.py --experiment_id 1590960724.3 --comment 'tf2-af03-t_s4-hyb32-tres2_gt_targ-fullstopg' --gt_targets --repeat_vae_sample --use_vae --kld_type kld_p0_norm --kld_weight 0.5 --kld_increment 0.99995 --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 4 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model transformer --transformer_scale --transformer_pos_encoding --transformer_layers 6 --transformer_heads 4 --transformer_dmodel 64 --transformer_hidden_units 256 --transformer_dropout 0.0 --latent_units 8 --decoder_model t_emb --decoder_dropout 0.0 --decoder_layers 4 --decoder_hidden_units 512,512,512,512 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position --disable_pen_loss --mask_encoder_pen",
"seq2seq-vae+cose_dec": "python ink_training_eager_predictive.py --experiment_id 1590960724.4 --comment 'tf2-af03-t_s4-hyb32-tres2-fullstopg' --repeat_vae_sample --use_vae --kld_type kld_p0_norm --kld_weight 0.5 --kld_increment 0.99995 --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 1 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model rnn --bidirectional_encoder --latent_units 8 --decoder_model rnn --decoder_dropout 0.0 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position",
"seq2seq-ar+cose_dec": "python ink_training_eager_predictive.py --experiment_id 1590960724.5 --comment 'tf2-af03-t_s4-hyb32-tres2-fullstopg' --decoder_autoregressive --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 1 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model rnn --bidirectional_encoder --latent_units 8 --decoder_model rnn --decoder_dropout 0.0 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position",
"seq2seq-vae-ar+cose_dec": "python ink_training_eager_predictive.py --experiment_id 1590960724.6 --comment 'tf2-af03-t_s4-hyb32-tres2-fullstopg' --decoder_autoregressive --repeat_vae_sample --use_vae --kld_type kld_p0_norm --kld_weight 0.5 --kld_increment 0.99995 --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 1 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model rnn --bidirectional_encoder --latent_units 8 --decoder_model rnn --decoder_dropout 0.0 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position",
"seq2seq+rnn": "python ink_training_eager_predictive.py --experiment_id 1591013267.2 --comment 'tf2-af03-ord32-tres2-fullstopg' --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type ordered --stroke_loss nll_gmm --n_t_samples 1 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model rnn --bidirectional_encoder --latent_units 8 --decoder_model rnn --decoder_dropout 0.0 --predictive_model rnn --learning_rate_type transformer --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model rnn --data_name didi_wo_text --metadata_type position",
"seq2seq+rnn (vel)": "python ink_training_eager_predictive.py --experiment_id 1591013267.1 --comment 'tf2-af03-ord32-tres2-fullstopg-vel' --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type ordered --stroke_loss nll_gmm --n_t_samples 1 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model rnn --bidirectional_encoder --latent_units 8 --decoder_model rnn --decoder_dropout 0.0 --predictive_model rnn --learning_rate_type exponential --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model rnn --data_name didi_wo_text --metadata_type velocity",
"seq2seq+cose": "python ink_training_eager_predictive.py --experiment_id 1597069595.3 --comment 'tf2-af03-t_s4-hyb32-tres2-fullstopg' --use_start_pos --num_pred_inputs 32 --stop_predictive_grad --pred_input_type hybrid --stroke_loss nll_gmm --n_t_samples 1 --batch_size 128 --affine_prob 0.3 --resampling_factor 2 --scale_factor 0 --grad_clip_norm 1 --encoder_model rnn --bidirectional_encoder --latent_units 8 --decoder_model rnn --decoder_dropout 0.0 --predictive_model transformer --learning_rate_type transformer --p_transformer_layers 6 --p_transformer_heads 4 --p_transformer_dmodel 64 --p_transformer_hidden_units 256 --p_transformer_dropout 0.0 --p_transformer_scale --embedding_loss nll_gmm --embedding_gmm_components 10 --loss_predicted_embedding --loss_reconstructed_ink --position_model transformer --data_name didi_wo_text --metadata_type position"
}