summaryrefslogtreecommitdiff
path: root/results
diff options
context:
space:
mode:
Diffstat (limited to 'results')
-rw-r--r--results/babyai_shared/b0/d2_lr0.01_bp.json151
-rw-r--r--results/babyai_shared/b0/d2_lr0.01_clean_kp.json151
-rw-r--r--results/babyai_shared/b0/d2_lr0.03_bp.json151
-rw-r--r--results/babyai_shared/b0/d2_lr0.03_clean_kp.json151
-rw-r--r--results/babyai_shared/b0/d4_lr0.01_bp.json151
-rw-r--r--results/babyai_shared/b0/d4_lr0.01_clean_kp.json151
-rw-r--r--results/babyai_shared/b0/d4_lr0.03_bp.json151
-rw-r--r--results/babyai_shared/b0/d4_lr0.03_clean_kp.json151
-rw-r--r--results/babyai_shared/b0_selector.json105
9 files changed, 1313 insertions, 0 deletions
diff --git a/results/babyai_shared/b0/d2_lr0.01_bp.json b/results/babyai_shared/b0/d2_lr0.01_bp.json
new file mode 100644
index 0000000..1e60df3
--- /dev/null
+++ b/results/babyai_shared/b0/d2_lr0.01_bp.json
@@ -0,0 +1,151 @@
+{
+ "condition": "bp",
+ "config": {
+ "action_dim": 7,
+ "context_gain": 1.0,
+ "hidden_layers": 2,
+ "input_dim": 984,
+ "learning_rate": 0.01,
+ "mission_dim": 13,
+ "momentum": 0.9,
+ "reciprocal_learning_rate": 0.01,
+ "weight_decay": 0.0001,
+ "width": 256
+ },
+ "data": {
+ "color_cardinality": 6,
+ "elapsed_seconds": 37.730633020401,
+ "env_id": "BabyAI-GoToObjS6-v1",
+ "minigrid_version": "3.1.0",
+ "object_cardinality": 11,
+ "protocol": "babyai_shared_feedback_b0",
+ "rollout_episodes": 500,
+ "rollout_seed_start": 200000,
+ "state_cardinality": 3,
+ "train_episodes": 20000,
+ "train_steps": 70159,
+ "validation_episodes": 2000,
+ "validation_steps": 7019,
+ "vocabulary": [
+ "<unk>",
+ "ball",
+ "blue",
+ "box",
+ "go",
+ "green",
+ "grey",
+ "key",
+ "purple",
+ "red",
+ "the",
+ "to",
+ "yellow"
+ ]
+ },
+ "epoch_history": [
+ {
+ "epoch": 1,
+ "train_loss": 0.35939537301659585
+ },
+ {
+ "epoch": 2,
+ "train_loss": 0.04168066847730767
+ },
+ {
+ "epoch": 3,
+ "train_loss": 0.01859026498753916
+ },
+ {
+ "epoch": 4,
+ "train_loss": 0.015395146521993659
+ },
+ {
+ "epoch": 5,
+ "train_loss": 0.015227786766534503
+ },
+ {
+ "epoch": 6,
+ "train_loss": 0.012962078500369734
+ },
+ {
+ "epoch": 7,
+ "train_loss": 0.013404185661241751
+ },
+ {
+ "epoch": 8,
+ "train_loss": 0.01239592205690728
+ },
+ {
+ "epoch": 9,
+ "train_loss": 0.012198219955157996
+ },
+ {
+ "epoch": 10,
+ "train_loss": 0.011407826012000442
+ },
+ {
+ "epoch": 11,
+ "train_loss": 0.010740509365939281
+ },
+ {
+ "epoch": 12,
+ "train_loss": 0.01174285890576853
+ },
+ {
+ "epoch": 13,
+ "train_loss": 0.012136201211251318
+ },
+ {
+ "epoch": 14,
+ "train_loss": 0.010752951244569637
+ },
+ {
+ "epoch": 15,
+ "train_loss": 0.010369447311809794
+ }
+ ],
+ "epochs_completed": 15,
+ "finite": true,
+ "first_nonfinite_epoch": null,
+ "mission_lesion_rollout": {
+ "episodes": 500,
+ "mean_length": 13.1,
+ "mean_return": 0.6429,
+ "success": 0.704
+ },
+ "mission_lesion_validation": {
+ "accuracy": 0.8139336087761789,
+ "loss": 0.5671851808553681
+ },
+ "predictor": [],
+ "provenance": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_peak_allocated_bytes": 39800832,
+ "cuda_visible_devices": "0",
+ "device": "cuda:0",
+ "git_commit": "20fd3577693593e3f62bcc4e1c2565dca5349944",
+ "git_dirty_tracked": false,
+ "minigrid_version": "3.1.0",
+ "torch_version": "2.3.1+cu118"
+ },
+ "rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "stage": "babyai_shared_b0",
+ "training": {
+ "batch_size": 256,
+ "model_seed": 4101,
+ "neutral_examples_per_epoch": 0,
+ "parameter_count_including_reciprocal_context_predictor": 394759,
+ "shuffle_seed": 4101,
+ "total_wall_seconds_including_rollouts": 72.15492010116577,
+ "training_wall_seconds": 68.66148805618286
+ },
+ "validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.010374719048008347
+ }
+}
diff --git a/results/babyai_shared/b0/d2_lr0.01_clean_kp.json b/results/babyai_shared/b0/d2_lr0.01_clean_kp.json
new file mode 100644
index 0000000..c63545b
--- /dev/null
+++ b/results/babyai_shared/b0/d2_lr0.01_clean_kp.json
@@ -0,0 +1,151 @@
+{
+ "condition": "clean_kp",
+ "config": {
+ "action_dim": 7,
+ "context_gain": 1.0,
+ "hidden_layers": 2,
+ "input_dim": 984,
+ "learning_rate": 0.01,
+ "mission_dim": 13,
+ "momentum": 0.9,
+ "reciprocal_learning_rate": 0.01,
+ "weight_decay": 0.0001,
+ "width": 256
+ },
+ "data": {
+ "color_cardinality": 6,
+ "elapsed_seconds": 37.730633020401,
+ "env_id": "BabyAI-GoToObjS6-v1",
+ "minigrid_version": "3.1.0",
+ "object_cardinality": 11,
+ "protocol": "babyai_shared_feedback_b0",
+ "rollout_episodes": 500,
+ "rollout_seed_start": 200000,
+ "state_cardinality": 3,
+ "train_episodes": 20000,
+ "train_steps": 70159,
+ "validation_episodes": 2000,
+ "validation_steps": 7019,
+ "vocabulary": [
+ "<unk>",
+ "ball",
+ "blue",
+ "box",
+ "go",
+ "green",
+ "grey",
+ "key",
+ "purple",
+ "red",
+ "the",
+ "to",
+ "yellow"
+ ]
+ },
+ "epoch_history": [
+ {
+ "epoch": 1,
+ "train_loss": 0.42220023420724
+ },
+ {
+ "epoch": 2,
+ "train_loss": 0.042795064787973056
+ },
+ {
+ "epoch": 3,
+ "train_loss": 0.01790234833363105
+ },
+ {
+ "epoch": 4,
+ "train_loss": 0.014970682340420106
+ },
+ {
+ "epoch": 5,
+ "train_loss": 0.014731389162638648
+ },
+ {
+ "epoch": 6,
+ "train_loss": 0.012619410693137482
+ },
+ {
+ "epoch": 7,
+ "train_loss": 0.013174676087642596
+ },
+ {
+ "epoch": 8,
+ "train_loss": 0.012047174987383186
+ },
+ {
+ "epoch": 9,
+ "train_loss": 0.011912513411900198
+ },
+ {
+ "epoch": 10,
+ "train_loss": 0.011198682525110516
+ },
+ {
+ "epoch": 11,
+ "train_loss": 0.010507440414618362
+ },
+ {
+ "epoch": 12,
+ "train_loss": 0.011464962014843795
+ },
+ {
+ "epoch": 13,
+ "train_loss": 0.011738500667905266
+ },
+ {
+ "epoch": 14,
+ "train_loss": 0.010477929167999802
+ },
+ {
+ "epoch": 15,
+ "train_loss": 0.010263334540861913
+ }
+ ],
+ "epochs_completed": 15,
+ "finite": true,
+ "first_nonfinite_epoch": null,
+ "mission_lesion_rollout": {
+ "episodes": 500,
+ "mean_length": 16.362,
+ "mean_return": 0.5511499999999999,
+ "success": 0.602
+ },
+ "mission_lesion_validation": {
+ "accuracy": 0.7602222538823195,
+ "loss": 0.7746106897906575
+ },
+ "predictor": [],
+ "provenance": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_peak_allocated_bytes": 39800832,
+ "cuda_visible_devices": "1",
+ "device": "cuda:0",
+ "git_commit": "20fd3577693593e3f62bcc4e1c2565dca5349944",
+ "git_dirty_tracked": false,
+ "minigrid_version": "3.1.0",
+ "torch_version": "2.3.1+cu118"
+ },
+ "rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "stage": "babyai_shared_b0",
+ "training": {
+ "batch_size": 256,
+ "model_seed": 4101,
+ "neutral_examples_per_epoch": 0,
+ "parameter_count_including_reciprocal_context_predictor": 394759,
+ "shuffle_seed": 4101,
+ "total_wall_seconds_including_rollouts": 72.49147534370422,
+ "training_wall_seconds": 68.89583253860474
+ },
+ "validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.010123796248337468
+ }
+}
diff --git a/results/babyai_shared/b0/d2_lr0.03_bp.json b/results/babyai_shared/b0/d2_lr0.03_bp.json
new file mode 100644
index 0000000..53c0cd2
--- /dev/null
+++ b/results/babyai_shared/b0/d2_lr0.03_bp.json
@@ -0,0 +1,151 @@
+{
+ "condition": "bp",
+ "config": {
+ "action_dim": 7,
+ "context_gain": 1.0,
+ "hidden_layers": 2,
+ "input_dim": 984,
+ "learning_rate": 0.03,
+ "mission_dim": 13,
+ "momentum": 0.9,
+ "reciprocal_learning_rate": 0.03,
+ "weight_decay": 0.0001,
+ "width": 256
+ },
+ "data": {
+ "color_cardinality": 6,
+ "elapsed_seconds": 37.730633020401,
+ "env_id": "BabyAI-GoToObjS6-v1",
+ "minigrid_version": "3.1.0",
+ "object_cardinality": 11,
+ "protocol": "babyai_shared_feedback_b0",
+ "rollout_episodes": 500,
+ "rollout_seed_start": 200000,
+ "state_cardinality": 3,
+ "train_episodes": 20000,
+ "train_steps": 70159,
+ "validation_episodes": 2000,
+ "validation_steps": 7019,
+ "vocabulary": [
+ "<unk>",
+ "ball",
+ "blue",
+ "box",
+ "go",
+ "green",
+ "grey",
+ "key",
+ "purple",
+ "red",
+ "the",
+ "to",
+ "yellow"
+ ]
+ },
+ "epoch_history": [
+ {
+ "epoch": 1,
+ "train_loss": 0.32622500834817236
+ },
+ {
+ "epoch": 2,
+ "train_loss": 0.022967058995907955
+ },
+ {
+ "epoch": 3,
+ "train_loss": 0.013182067105292596
+ },
+ {
+ "epoch": 4,
+ "train_loss": 0.12631320300893012
+ },
+ {
+ "epoch": 5,
+ "train_loss": 0.03265200278869915
+ },
+ {
+ "epoch": 6,
+ "train_loss": 0.01237200420434502
+ },
+ {
+ "epoch": 7,
+ "train_loss": 0.012444034099012655
+ },
+ {
+ "epoch": 8,
+ "train_loss": 0.010830031347418712
+ },
+ {
+ "epoch": 9,
+ "train_loss": 0.010618501521848057
+ },
+ {
+ "epoch": 10,
+ "train_loss": 0.010414972044315867
+ },
+ {
+ "epoch": 11,
+ "train_loss": 0.009751175390768118
+ },
+ {
+ "epoch": 12,
+ "train_loss": 0.01019353245279159
+ },
+ {
+ "epoch": 13,
+ "train_loss": 0.010311937543394213
+ },
+ {
+ "epoch": 14,
+ "train_loss": 0.00966560988549397
+ },
+ {
+ "epoch": 15,
+ "train_loss": 0.009808915172311986
+ }
+ ],
+ "epochs_completed": 15,
+ "finite": true,
+ "first_nonfinite_epoch": null,
+ "mission_lesion_rollout": {
+ "episodes": 500,
+ "mean_length": 6.608,
+ "mean_return": 0.8254000000000001,
+ "success": 0.906
+ },
+ "mission_lesion_validation": {
+ "accuracy": 0.9393075936743126,
+ "loss": 0.15966886549462006
+ },
+ "predictor": [],
+ "provenance": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_peak_allocated_bytes": 39800832,
+ "cuda_visible_devices": "2",
+ "device": "cuda:0",
+ "git_commit": "20fd3577693593e3f62bcc4e1c2565dca5349944",
+ "git_dirty_tracked": false,
+ "minigrid_version": "3.1.0",
+ "torch_version": "2.3.1+cu118"
+ },
+ "rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "stage": "babyai_shared_b0",
+ "training": {
+ "batch_size": 256,
+ "model_seed": 4101,
+ "neutral_examples_per_epoch": 0,
+ "parameter_count_including_reciprocal_context_predictor": 394759,
+ "shuffle_seed": 4101,
+ "total_wall_seconds_including_rollouts": 71.35738158226013,
+ "training_wall_seconds": 68.60933923721313
+ },
+ "validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.008090585890548625
+ }
+}
diff --git a/results/babyai_shared/b0/d2_lr0.03_clean_kp.json b/results/babyai_shared/b0/d2_lr0.03_clean_kp.json
new file mode 100644
index 0000000..767f65d
--- /dev/null
+++ b/results/babyai_shared/b0/d2_lr0.03_clean_kp.json
@@ -0,0 +1,151 @@
+{
+ "condition": "clean_kp",
+ "config": {
+ "action_dim": 7,
+ "context_gain": 1.0,
+ "hidden_layers": 2,
+ "input_dim": 984,
+ "learning_rate": 0.03,
+ "mission_dim": 13,
+ "momentum": 0.9,
+ "reciprocal_learning_rate": 0.03,
+ "weight_decay": 0.0001,
+ "width": 256
+ },
+ "data": {
+ "color_cardinality": 6,
+ "elapsed_seconds": 37.730633020401,
+ "env_id": "BabyAI-GoToObjS6-v1",
+ "minigrid_version": "3.1.0",
+ "object_cardinality": 11,
+ "protocol": "babyai_shared_feedback_b0",
+ "rollout_episodes": 500,
+ "rollout_seed_start": 200000,
+ "state_cardinality": 3,
+ "train_episodes": 20000,
+ "train_steps": 70159,
+ "validation_episodes": 2000,
+ "validation_steps": 7019,
+ "vocabulary": [
+ "<unk>",
+ "ball",
+ "blue",
+ "box",
+ "go",
+ "green",
+ "grey",
+ "key",
+ "purple",
+ "red",
+ "the",
+ "to",
+ "yellow"
+ ]
+ },
+ "epoch_history": [
+ {
+ "epoch": 1,
+ "train_loss": 0.4284296942739324
+ },
+ {
+ "epoch": 2,
+ "train_loss": 0.021993979699909686
+ },
+ {
+ "epoch": 3,
+ "train_loss": 0.012347337700511244
+ },
+ {
+ "epoch": 4,
+ "train_loss": 0.11120144955475222
+ },
+ {
+ "epoch": 5,
+ "train_loss": 0.02621290213089775
+ },
+ {
+ "epoch": 6,
+ "train_loss": 0.011413470651869748
+ },
+ {
+ "epoch": 7,
+ "train_loss": 0.011871577144404661
+ },
+ {
+ "epoch": 8,
+ "train_loss": 0.010163592805302787
+ },
+ {
+ "epoch": 9,
+ "train_loss": 0.009837346972728317
+ },
+ {
+ "epoch": 10,
+ "train_loss": 0.01011049080259082
+ },
+ {
+ "epoch": 11,
+ "train_loss": 0.009574851958419789
+ },
+ {
+ "epoch": 12,
+ "train_loss": 0.009629912125899202
+ },
+ {
+ "epoch": 13,
+ "train_loss": 0.009786839098352092
+ },
+ {
+ "epoch": 14,
+ "train_loss": 0.009409503974602558
+ },
+ {
+ "epoch": 15,
+ "train_loss": 0.009612279576884413
+ }
+ ],
+ "epochs_completed": 15,
+ "finite": true,
+ "first_nonfinite_epoch": null,
+ "mission_lesion_rollout": {
+ "episodes": 500,
+ "mean_length": 5.856,
+ "mean_return": 0.8466,
+ "success": 0.93
+ },
+ "mission_lesion_validation": {
+ "accuracy": 0.9532696965379683,
+ "loss": 0.12229915629222635
+ },
+ "predictor": [],
+ "provenance": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_peak_allocated_bytes": 39800832,
+ "cuda_visible_devices": "3",
+ "device": "cuda:0",
+ "git_commit": "20fd3577693593e3f62bcc4e1c2565dca5349944",
+ "git_dirty_tracked": false,
+ "minigrid_version": "3.1.0",
+ "torch_version": "2.3.1+cu118"
+ },
+ "rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "stage": "babyai_shared_b0",
+ "training": {
+ "batch_size": 256,
+ "model_seed": 4101,
+ "neutral_examples_per_epoch": 0,
+ "parameter_count_including_reciprocal_context_predictor": 394759,
+ "shuffle_seed": 4101,
+ "total_wall_seconds_including_rollouts": 71.55634665489197,
+ "training_wall_seconds": 68.94226169586182
+ },
+ "validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.007343835429191453
+ }
+}
diff --git a/results/babyai_shared/b0/d4_lr0.01_bp.json b/results/babyai_shared/b0/d4_lr0.01_bp.json
new file mode 100644
index 0000000..ce30242
--- /dev/null
+++ b/results/babyai_shared/b0/d4_lr0.01_bp.json
@@ -0,0 +1,151 @@
+{
+ "condition": "bp",
+ "config": {
+ "action_dim": 7,
+ "context_gain": 1.0,
+ "hidden_layers": 4,
+ "input_dim": 984,
+ "learning_rate": 0.01,
+ "mission_dim": 13,
+ "momentum": 0.9,
+ "reciprocal_learning_rate": 0.01,
+ "weight_decay": 0.0001,
+ "width": 256
+ },
+ "data": {
+ "color_cardinality": 6,
+ "elapsed_seconds": 37.730633020401,
+ "env_id": "BabyAI-GoToObjS6-v1",
+ "minigrid_version": "3.1.0",
+ "object_cardinality": 11,
+ "protocol": "babyai_shared_feedback_b0",
+ "rollout_episodes": 500,
+ "rollout_seed_start": 200000,
+ "state_cardinality": 3,
+ "train_episodes": 20000,
+ "train_steps": 70159,
+ "validation_episodes": 2000,
+ "validation_steps": 7019,
+ "vocabulary": [
+ "<unk>",
+ "ball",
+ "blue",
+ "box",
+ "go",
+ "green",
+ "grey",
+ "key",
+ "purple",
+ "red",
+ "the",
+ "to",
+ "yellow"
+ ]
+ },
+ "epoch_history": [
+ {
+ "epoch": 1,
+ "train_loss": 0.3253778777474707
+ },
+ {
+ "epoch": 2,
+ "train_loss": 0.019206965685060078
+ },
+ {
+ "epoch": 3,
+ "train_loss": 0.012563892805271528
+ },
+ {
+ "epoch": 4,
+ "train_loss": 0.013036079497330568
+ },
+ {
+ "epoch": 5,
+ "train_loss": 0.014585806006692688
+ },
+ {
+ "epoch": 6,
+ "train_loss": 0.011469631964256141
+ },
+ {
+ "epoch": 7,
+ "train_loss": 0.012590554792894346
+ },
+ {
+ "epoch": 8,
+ "train_loss": 0.010635647804857316
+ },
+ {
+ "epoch": 9,
+ "train_loss": 0.01044608271402963
+ },
+ {
+ "epoch": 10,
+ "train_loss": 0.01032080053330653
+ },
+ {
+ "epoch": 11,
+ "train_loss": 0.009672622701830485
+ },
+ {
+ "epoch": 12,
+ "train_loss": 0.010132916131353175
+ },
+ {
+ "epoch": 13,
+ "train_loss": 0.010232909059956332
+ },
+ {
+ "epoch": 14,
+ "train_loss": 0.009590742462103001
+ },
+ {
+ "epoch": 15,
+ "train_loss": 0.009766413487642157
+ }
+ ],
+ "epochs_completed": 15,
+ "finite": true,
+ "first_nonfinite_epoch": null,
+ "mission_lesion_rollout": {
+ "episodes": 500,
+ "mean_length": 18.046,
+ "mean_return": 0.50385,
+ "success": 0.55
+ },
+ "mission_lesion_validation": {
+ "accuracy": 0.7350049864653084,
+ "loss": 1.0839614618162532
+ },
+ "predictor": [],
+ "provenance": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_peak_allocated_bytes": 42582016,
+ "cuda_visible_devices": "4",
+ "device": "cuda:0",
+ "git_commit": "20fd3577693593e3f62bcc4e1c2565dca5349944",
+ "git_dirty_tracked": false,
+ "minigrid_version": "3.1.0",
+ "torch_version": "2.3.1+cu118"
+ },
+ "rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "stage": "babyai_shared_b0",
+ "training": {
+ "batch_size": 256,
+ "model_seed": 4101,
+ "neutral_examples_per_epoch": 0,
+ "parameter_count_including_reciprocal_context_predictor": 665095,
+ "shuffle_seed": 4101,
+ "total_wall_seconds_including_rollouts": 69.9526629447937,
+ "training_wall_seconds": 63.36152243614197
+ },
+ "validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.007651511259713494
+ }
+}
diff --git a/results/babyai_shared/b0/d4_lr0.01_clean_kp.json b/results/babyai_shared/b0/d4_lr0.01_clean_kp.json
new file mode 100644
index 0000000..72a132c
--- /dev/null
+++ b/results/babyai_shared/b0/d4_lr0.01_clean_kp.json
@@ -0,0 +1,151 @@
+{
+ "condition": "clean_kp",
+ "config": {
+ "action_dim": 7,
+ "context_gain": 1.0,
+ "hidden_layers": 4,
+ "input_dim": 984,
+ "learning_rate": 0.01,
+ "mission_dim": 13,
+ "momentum": 0.9,
+ "reciprocal_learning_rate": 0.01,
+ "weight_decay": 0.0001,
+ "width": 256
+ },
+ "data": {
+ "color_cardinality": 6,
+ "elapsed_seconds": 37.730633020401,
+ "env_id": "BabyAI-GoToObjS6-v1",
+ "minigrid_version": "3.1.0",
+ "object_cardinality": 11,
+ "protocol": "babyai_shared_feedback_b0",
+ "rollout_episodes": 500,
+ "rollout_seed_start": 200000,
+ "state_cardinality": 3,
+ "train_episodes": 20000,
+ "train_steps": 70159,
+ "validation_episodes": 2000,
+ "validation_steps": 7019,
+ "vocabulary": [
+ "<unk>",
+ "ball",
+ "blue",
+ "box",
+ "go",
+ "green",
+ "grey",
+ "key",
+ "purple",
+ "red",
+ "the",
+ "to",
+ "yellow"
+ ]
+ },
+ "epoch_history": [
+ {
+ "epoch": 1,
+ "train_loss": 0.4667839081585407
+ },
+ {
+ "epoch": 2,
+ "train_loss": 0.04674517553976991
+ },
+ {
+ "epoch": 3,
+ "train_loss": 0.01326043784279715
+ },
+ {
+ "epoch": 4,
+ "train_loss": 0.061285746785896746
+ },
+ {
+ "epoch": 5,
+ "train_loss": 0.026032158348878676
+ },
+ {
+ "epoch": 6,
+ "train_loss": 0.013545795751904899
+ },
+ {
+ "epoch": 7,
+ "train_loss": 0.014207788907065564
+ },
+ {
+ "epoch": 8,
+ "train_loss": 0.011456064975143156
+ },
+ {
+ "epoch": 9,
+ "train_loss": 0.011107088818402685
+ },
+ {
+ "epoch": 10,
+ "train_loss": 0.010615441516486251
+ },
+ {
+ "epoch": 11,
+ "train_loss": 0.009885513414840468
+ },
+ {
+ "epoch": 12,
+ "train_loss": 0.01040836228650402
+ },
+ {
+ "epoch": 13,
+ "train_loss": 0.010366036330180411
+ },
+ {
+ "epoch": 14,
+ "train_loss": 0.009639627484350719
+ },
+ {
+ "epoch": 15,
+ "train_loss": 0.009855725453976033
+ }
+ ],
+ "epochs_completed": 15,
+ "finite": true,
+ "first_nonfinite_epoch": null,
+ "mission_lesion_rollout": {
+ "episodes": 500,
+ "mean_length": 4.292,
+ "mean_return": 0.8905000000000001,
+ "success": 0.978
+ },
+ "mission_lesion_validation": {
+ "accuracy": 0.9955834164410885,
+ "loss": 0.01790462883301042
+ },
+ "predictor": [],
+ "provenance": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_peak_allocated_bytes": 42582016,
+ "cuda_visible_devices": "5",
+ "device": "cuda:0",
+ "git_commit": "20fd3577693593e3f62bcc4e1c2565dca5349944",
+ "git_dirty_tracked": false,
+ "minigrid_version": "3.1.0",
+ "torch_version": "2.3.1+cu118"
+ },
+ "rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "stage": "babyai_shared_b0",
+ "training": {
+ "batch_size": 256,
+ "model_seed": 4101,
+ "neutral_examples_per_epoch": 0,
+ "parameter_count_including_reciprocal_context_predictor": 665095,
+ "shuffle_seed": 4101,
+ "total_wall_seconds_including_rollouts": 68.48475790023804,
+ "training_wall_seconds": 64.21018767356873
+ },
+ "validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.007562428802009797
+ }
+}
diff --git a/results/babyai_shared/b0/d4_lr0.03_bp.json b/results/babyai_shared/b0/d4_lr0.03_bp.json
new file mode 100644
index 0000000..33f8418
--- /dev/null
+++ b/results/babyai_shared/b0/d4_lr0.03_bp.json
@@ -0,0 +1,151 @@
+{
+ "condition": "bp",
+ "config": {
+ "action_dim": 7,
+ "context_gain": 1.0,
+ "hidden_layers": 4,
+ "input_dim": 984,
+ "learning_rate": 0.03,
+ "mission_dim": 13,
+ "momentum": 0.9,
+ "reciprocal_learning_rate": 0.03,
+ "weight_decay": 0.0001,
+ "width": 256
+ },
+ "data": {
+ "color_cardinality": 6,
+ "elapsed_seconds": 37.730633020401,
+ "env_id": "BabyAI-GoToObjS6-v1",
+ "minigrid_version": "3.1.0",
+ "object_cardinality": 11,
+ "protocol": "babyai_shared_feedback_b0",
+ "rollout_episodes": 500,
+ "rollout_seed_start": 200000,
+ "state_cardinality": 3,
+ "train_episodes": 20000,
+ "train_steps": 70159,
+ "validation_episodes": 2000,
+ "validation_steps": 7019,
+ "vocabulary": [
+ "<unk>",
+ "ball",
+ "blue",
+ "box",
+ "go",
+ "green",
+ "grey",
+ "key",
+ "purple",
+ "red",
+ "the",
+ "to",
+ "yellow"
+ ]
+ },
+ "epoch_history": [
+ {
+ "epoch": 1,
+ "train_loss": 0.5505747761509635
+ },
+ {
+ "epoch": 2,
+ "train_loss": 0.3316267966343598
+ },
+ {
+ "epoch": 3,
+ "train_loss": 0.09943052434563553
+ },
+ {
+ "epoch": 4,
+ "train_loss": 0.04950741482962092
+ },
+ {
+ "epoch": 5,
+ "train_loss": 0.013397140696906718
+ },
+ {
+ "epoch": 6,
+ "train_loss": 0.010083013286560097
+ },
+ {
+ "epoch": 7,
+ "train_loss": 0.011028704402918017
+ },
+ {
+ "epoch": 8,
+ "train_loss": 0.009480555155737834
+ },
+ {
+ "epoch": 9,
+ "train_loss": 0.009246192916672788
+ },
+ {
+ "epoch": 10,
+ "train_loss": 0.009409724382448158
+ },
+ {
+ "epoch": 11,
+ "train_loss": 0.00913212353257801
+ },
+ {
+ "epoch": 12,
+ "train_loss": 0.00898012409410016
+ },
+ {
+ "epoch": 13,
+ "train_loss": 0.009049554331206971
+ },
+ {
+ "epoch": 14,
+ "train_loss": 0.00897477681505155
+ },
+ {
+ "epoch": 15,
+ "train_loss": 0.009097340858795426
+ }
+ ],
+ "epochs_completed": 15,
+ "finite": true,
+ "first_nonfinite_epoch": null,
+ "mission_lesion_rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "mission_lesion_validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.009770111168767503
+ },
+ "predictor": [],
+ "provenance": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_peak_allocated_bytes": 42582016,
+ "cuda_visible_devices": "6",
+ "device": "cuda:0",
+ "git_commit": "20fd3577693593e3f62bcc4e1c2565dca5349944",
+ "git_dirty_tracked": false,
+ "minigrid_version": "3.1.0",
+ "torch_version": "2.3.1+cu118"
+ },
+ "rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "stage": "babyai_shared_b0",
+ "training": {
+ "batch_size": 256,
+ "model_seed": 4101,
+ "neutral_examples_per_epoch": 0,
+ "parameter_count_including_reciprocal_context_predictor": 665095,
+ "shuffle_seed": 4101,
+ "total_wall_seconds_including_rollouts": 67.55784368515015,
+ "training_wall_seconds": 63.23557496070862
+ },
+ "validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.00700099701928284
+ }
+}
diff --git a/results/babyai_shared/b0/d4_lr0.03_clean_kp.json b/results/babyai_shared/b0/d4_lr0.03_clean_kp.json
new file mode 100644
index 0000000..b0c853d
--- /dev/null
+++ b/results/babyai_shared/b0/d4_lr0.03_clean_kp.json
@@ -0,0 +1,151 @@
+{
+ "condition": "clean_kp",
+ "config": {
+ "action_dim": 7,
+ "context_gain": 1.0,
+ "hidden_layers": 4,
+ "input_dim": 984,
+ "learning_rate": 0.03,
+ "mission_dim": 13,
+ "momentum": 0.9,
+ "reciprocal_learning_rate": 0.03,
+ "weight_decay": 0.0001,
+ "width": 256
+ },
+ "data": {
+ "color_cardinality": 6,
+ "elapsed_seconds": 37.730633020401,
+ "env_id": "BabyAI-GoToObjS6-v1",
+ "minigrid_version": "3.1.0",
+ "object_cardinality": 11,
+ "protocol": "babyai_shared_feedback_b0",
+ "rollout_episodes": 500,
+ "rollout_seed_start": 200000,
+ "state_cardinality": 3,
+ "train_episodes": 20000,
+ "train_steps": 70159,
+ "validation_episodes": 2000,
+ "validation_steps": 7019,
+ "vocabulary": [
+ "<unk>",
+ "ball",
+ "blue",
+ "box",
+ "go",
+ "green",
+ "grey",
+ "key",
+ "purple",
+ "red",
+ "the",
+ "to",
+ "yellow"
+ ]
+ },
+ "epoch_history": [
+ {
+ "epoch": 1,
+ "train_loss": 0.560454016964544
+ },
+ {
+ "epoch": 2,
+ "train_loss": 0.11413068584420465
+ },
+ {
+ "epoch": 3,
+ "train_loss": 0.01235340728620279
+ },
+ {
+ "epoch": 4,
+ "train_loss": 0.010933047032872723
+ },
+ {
+ "epoch": 5,
+ "train_loss": 0.010575457136702342
+ },
+ {
+ "epoch": 6,
+ "train_loss": 0.00985285716623449
+ },
+ {
+ "epoch": 7,
+ "train_loss": 0.010482580381583168
+ },
+ {
+ "epoch": 8,
+ "train_loss": 0.009581763261918572
+ },
+ {
+ "epoch": 9,
+ "train_loss": 0.009271365735324005
+ },
+ {
+ "epoch": 10,
+ "train_loss": 0.00958018275731857
+ },
+ {
+ "epoch": 11,
+ "train_loss": 0.009324251855247316
+ },
+ {
+ "epoch": 12,
+ "train_loss": 0.00918881041791544
+ },
+ {
+ "epoch": 13,
+ "train_loss": 0.009378983245802705
+ },
+ {
+ "epoch": 14,
+ "train_loss": 0.009324983275675384
+ },
+ {
+ "epoch": 15,
+ "train_loss": 0.009338655431241601
+ }
+ ],
+ "epochs_completed": 15,
+ "finite": true,
+ "first_nonfinite_epoch": null,
+ "mission_lesion_rollout": {
+ "episodes": 500,
+ "mean_length": 4.306,
+ "mean_return": 0.8901500000000001,
+ "success": 0.978
+ },
+ "mission_lesion_validation": {
+ "accuracy": 0.9921641259438666,
+ "loss": 0.029844395743083507
+ },
+ "predictor": [],
+ "provenance": {
+ "cuda_device_name": "NVIDIA GeForce GTX 1080",
+ "cuda_peak_allocated_bytes": 42582016,
+ "cuda_visible_devices": "7",
+ "device": "cuda:0",
+ "git_commit": "20fd3577693593e3f62bcc4e1c2565dca5349944",
+ "git_dirty_tracked": false,
+ "minigrid_version": "3.1.0",
+ "torch_version": "2.3.1+cu118"
+ },
+ "rollout": {
+ "episodes": 500,
+ "mean_length": 4.29,
+ "mean_return": 0.8905500000000001,
+ "success": 0.978
+ },
+ "stage": "babyai_shared_b0",
+ "training": {
+ "batch_size": 256,
+ "model_seed": 4101,
+ "neutral_examples_per_epoch": 0,
+ "parameter_count_including_reciprocal_context_predictor": 665095,
+ "shuffle_seed": 4101,
+ "total_wall_seconds_including_rollouts": 72.05698609352112,
+ "training_wall_seconds": 69.7202775478363
+ },
+ "validation": {
+ "accuracy": 0.9974355321270836,
+ "loss": 0.006933236702977295
+ }
+}
diff --git a/results/babyai_shared/b0_selector.json b/results/babyai_shared/b0_selector.json
new file mode 100644
index 0000000..4368e44
--- /dev/null
+++ b/results/babyai_shared/b0_selector.json
@@ -0,0 +1,105 @@
+{
+ "candidates": [
+ {
+ "bp_action_accuracy": 0.9974355321270836,
+ "bp_mission_lesion_drop": 0.274,
+ "bp_mission_lesion_success": 0.704,
+ "bp_rollout_success": 0.978,
+ "checks": {
+ "both_finite": true,
+ "bp_mission_lesion_drop_at_least_0p2": true,
+ "bp_success_at_least_0p8": true,
+ "clean_kp_success_at_least_0p8": true
+ },
+ "clean_kp_action_accuracy": 0.9974355321270836,
+ "clean_kp_rollout_success": 0.978,
+ "eligible": true,
+ "hidden_layers": 2,
+ "learning_rate": 0.01,
+ "source_files": [
+ "results/babyai_shared/b0/d2_lr0.01_bp.json",
+ "results/babyai_shared/b0/d2_lr0.01_clean_kp.json"
+ ]
+ },
+ {
+ "bp_action_accuracy": 0.9974355321270836,
+ "bp_mission_lesion_drop": 0.07199999999999995,
+ "bp_mission_lesion_success": 0.906,
+ "bp_rollout_success": 0.978,
+ "checks": {
+ "both_finite": true,
+ "bp_mission_lesion_drop_at_least_0p2": false,
+ "bp_success_at_least_0p8": true,
+ "clean_kp_success_at_least_0p8": true
+ },
+ "clean_kp_action_accuracy": 0.9974355321270836,
+ "clean_kp_rollout_success": 0.978,
+ "eligible": false,
+ "hidden_layers": 2,
+ "learning_rate": 0.03,
+ "source_files": [
+ "results/babyai_shared/b0/d2_lr0.03_bp.json",
+ "results/babyai_shared/b0/d2_lr0.03_clean_kp.json"
+ ]
+ },
+ {
+ "bp_action_accuracy": 0.9974355321270836,
+ "bp_mission_lesion_drop": 0.42799999999999994,
+ "bp_mission_lesion_success": 0.55,
+ "bp_rollout_success": 0.978,
+ "checks": {
+ "both_finite": true,
+ "bp_mission_lesion_drop_at_least_0p2": true,
+ "bp_success_at_least_0p8": true,
+ "clean_kp_success_at_least_0p8": true
+ },
+ "clean_kp_action_accuracy": 0.9974355321270836,
+ "clean_kp_rollout_success": 0.978,
+ "eligible": true,
+ "hidden_layers": 4,
+ "learning_rate": 0.01,
+ "source_files": [
+ "results/babyai_shared/b0/d4_lr0.01_bp.json",
+ "results/babyai_shared/b0/d4_lr0.01_clean_kp.json"
+ ]
+ },
+ {
+ "bp_action_accuracy": 0.9974355321270836,
+ "bp_mission_lesion_drop": 0.0,
+ "bp_mission_lesion_success": 0.978,
+ "bp_rollout_success": 0.978,
+ "checks": {
+ "both_finite": true,
+ "bp_mission_lesion_drop_at_least_0p2": false,
+ "bp_success_at_least_0p8": true,
+ "clean_kp_success_at_least_0p8": true
+ },
+ "clean_kp_action_accuracy": 0.9974355321270836,
+ "clean_kp_rollout_success": 0.978,
+ "eligible": false,
+ "hidden_layers": 4,
+ "learning_rate": 0.03,
+ "source_files": [
+ "results/babyai_shared/b0/d4_lr0.03_bp.json",
+ "results/babyai_shared/b0/d4_lr0.03_clean_kp.json"
+ ]
+ }
+ ],
+ "gate": "pass",
+ "raw_or_sdil_results_read": false,
+ "selected": {
+ "b1_epochs": 40,
+ "b1_model_and_shuffle_seeds": [
+ 4101,
+ 4102,
+ 4103
+ ],
+ "context_gain": 1.0,
+ "hidden_layers": 2,
+ "learning_rate": 0.01,
+ "width": 256
+ },
+ "selection_rule": "highest clean-KP rollout success, then action accuracy, then fewer layers, then smaller learning rate, among eligible rows",
+ "stage": "babyai_shared_b0_selector",
+ "test_split_generated_or_read": false
+}
generated by cgit v1.2.3 (git 2.25.1) at 2026-09-15 08:45:43 +0000