KAN-24: display heart 25

.gitignore

@@ -1,2 +1,3 @@
 logs/
-results/
+prj/
+__pycache__/

config.yaml

@@ -1,14 +1,17 @@
 ---
+
 experiment:
   seed: 42
-  save_dir: ./results
-  save_model: model
-  gpus: [0]
+  save_dir: ./prj/exp_mnist_bayes_lenet_try
+  save_model: mnist_bayes_lenet_10samples_mc
+  cpu: True
   ckpt_file: best_chkp.tf
 
 dataset:
   name: mnist
   data: <dict>
+  mean: 0.0
+  std: 1.0
 
 model:
   original: <obj>
@@ -21,54 +24,93 @@ model:
   dropout_type: mc
   num_bayes_layer: 3
 
+# NOTE: introduce typing so that we can type dictionaries too
+S1:
+  accuracy: 
+    weight: 0.25
+    base: 0.99
+  ece: 
+    weight: -0.25
+    base: 0.03
+  ape: 
+    weight: 0.25
+    base: 1.5
+  flops: 
+    weight: -0.25
+    base: 5340192
+
+  top_n: 3
+  # special metric calculated from the other metrics
+  top_metric: score
+
+  name: Balance
+
+S2:
+  accuracy: 
+    weight: 0.75
+    base: 0.99
+  ece: 
+    weight: 0.00
+    base: 0.03
+  ape: 
+    weight: 0.00
+    base: 1.5
+  flops: 
+    weight: -0.25
+    base: 5340192
+
+  top_n: 1
+  top_metric: flops
+
+  name: Opt-Efficiency
+
+strategy:
+  strategies: [ S1 ]
+  curr_strategy: <int>
+  curr_results: ((bayes_opt.summary))
+
+  terminate_strategies: <bool>
+  results: <dict>
+  save_dir: ((experiment.save_dir))
+
 
 bayes_opt:
   iteration: <int>
   terminate: <bool>
   engine: <obj>
   score: <float>
   summary: <list>
+
+  curr_strategy: <obj>
+
   control:
     params: dict
-    metrics: dict
     suggests: dict
 
-  num_iter: 3
+  num_iter: 2
   seed: ((experiment.seed))
   tunable:
     dropout_rate:
        value: ((model.dropout_rate))
-       space: [0.1, 0.2, 0.3, 0.4]
+       space: [0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.80, 0.85, 0.90, 0.95]
+      #  space: [0.1, 0.2, 0.3, 0.4]
     p_rate:
        value: ((model.p_rate))
-       space:  [0.0, 0.1, 0.2]
+       space:  [0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.80, 0.85, 0.90, 0.95]
+      #  space: [0.0, 0.1, 0.2]
     num_bayes_layer:
        value: ((model.num_bayes_layer))
        space: [1, 2, 3]
     scale_factor:
        value: ((model.scale_factor))
-       space: [0.5, 1.0, 1.5]
+      #  space: [0.5, 1.0, 1.5]
+       space: [0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.80, 0.85, 0.90, 0.95]
 
   metrics:
     accuracy: ((eval.accuracy))
     ece: ((eval.ece))
-    aPE: ((eval.ape))
-    FLOP: ((eval.flop))
-
-  score_weights:
-    accuracy:
-      weight: 0.25
-      base: 0.99
-    ece:
-      weight: 0.25
-      base: -0.03
-    aPE:
-      weight: 0.25
-      base: 1.5
-    FLOP:
-      weight: 0.25
-      base: -5340192
-
+    ape: ((eval.ape))
+    flops: ((eval.flops))
 
 train:
   optimizer: <obj>
@@ -82,7 +124,7 @@ eval:
   ece: <float>
   ape: <float>
   accuracy: <float>
-  flop: <float>
+  flops: <float>
 
   mc_samples: 5
   num_eval_images: 200

config_resnet.yaml

@@ -0,0 +1,145 @@
+---
+
+experiment:
+  seed: 42
+  save_dir: ./prj/exp_cifar_bayes_resnet
+  save_model: cifar_resnet_tmp_3bayeslayer_mc
+  gpus: [0, 1]
+  ckpt_file: best_chkp.tf
+
+dataset:
+  name: cifar10
+  data: <dict>
+  mean: 0.0
+  std: 1.0
+
+model:
+  original: <obj>
+  logic: <obj>
+  name: resnet
+  is_quant: false
+  dropout_rate: 0.4
+  p_rate: 0.05
+  scale_factor: 0.3
+  dropout_type: mc
+  num_bayes_layer: 3
+
+# NOTE: introduce typing so that we can type dictionaries too
+S1:
+  accuracy: 
+    weight: 0.25
+    base: 0.99
+  ece: 
+    weight: -0.25
+    base: 0.03
+  ape: 
+    weight: 0.25
+    base: 1.5
+  flops: 
+    weight: -0.25
+    base: 5340192
+
+  top_n: 3
+  # special metric calculated from the other metrics
+  top_metric: score
+
+  name: Balance
+
+S2:
+  accuracy: 
+    weight: 0.75
+    base: 0.99
+  ece: 
+    weight: 0.00
+    base: 0.03
+  ape: 
+    weight: 0.00
+    base: 1.5
+  flops: 
+    weight: -0.25
+    base: 5340192
+
+  top_n: 1
+  top_metric: flops
+
+  name: Opt-Efficiency
+
+strategy:
+  strategies: [ S1 ]
+  curr_strategy: <int>
+  curr_results: ((bayes_opt.summary))
+
+  terminate_strategies: <bool>
+  results: <dict>
+  save_dir: ((experiment.save_dir))
+
+
+bayes_opt:
+  iteration: <int>
+  terminate: <bool>
+  engine: <obj>
+  score: <float>
+  summary: <list>
+
+  curr_strategy: <obj>
+
+  control:
+    params: dict
+    suggests: dict
+
+  num_iter: 2
+  seed: ((experiment.seed))
+  tunable:
+    dropout_rate:
+       value: ((model.dropout_rate))
+       space: [0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.80, 0.85, 0.90, 0.95]
+      #  space: [0.1, 0.2, 0.3, 0.4]
+    p_rate:
+       value: ((model.p_rate))
+       space:  [0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.80, 0.85, 0.90, 0.95]
+      #  space: [0.0, 0.1, 0.2]
+    num_bayes_layer:
+       value: ((model.num_bayes_layer))
+       space: [1, 2, 3]
+    scale_factor:
+       value: ((model.scale_factor))
+      #  space: [0.5, 1.0, 1.5]
+       space: [0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.80, 0.85, 0.90, 0.95]
+
+  metrics:
+    accuracy: ((eval.accuracy))
+    ece: ((eval.ece))
+    ape: ((eval.ape))
+    flops: ((eval.flops))
+
+train:
+  optimizer: <obj>
+  id: ((bayes_opt.iteration))
+  num_epoch: 3
+  batch_size: 128
+  learning_rate: 0.01
+  validation_split: 0.1
+
+eval:
+  ece: <float>
+  ape: <float>
+  accuracy: <float>
+  flops: <float>
+
+  mc_samples: 10
+  num_eval_images: 200
+  num_bins: 10
+
+
+reporter:
+  log:
+    - bayes_opt.iteration
+    - dropout_rate_list
+    - p_rate
+    - num_bayes_layer
+    - scale_factor
+    - accuracy
+    - flops
+    - ece
+    - ape
+    - score

environment.yml

@@ -10,6 +10,9 @@ dependencies:
   - tensorflow=2.12
   - nvidia/label/cuda-11.8.0::cuda-nvcc
   - nvidia/label/cuda-11.8.0::cuda-toolkit
+  - matplotlib=3.7
+  - seaborn=0.13.2
+  - numpy<1.24
   - pip
   - pip:
       - -r requirements.txt

logic/pipeline.py

@@ -175,14 +175,6 @@ def on_epoch_end(self, epoch, logs=None):
 
         callbacks.append(PrintEpochCallback())
 
-
-
-        logger.info(f"p_rate: {cfg.model.p_rate}")
-        logger.info(f"batch_size: {cfg.training.batch_size}")
-        logger.info(f"num_epoch: {cfg.training.num_epoch}")
-        logger.debug(f"validation_split: {cfg.training.validation_split}")
-
-
         def hash_model(model):
             import hashlib
             import json

nautic/config.py

@@ -0,0 +1,11 @@
+# Config structure for the flowx
+class FlowxConfig:
+    disable_nautic: bool = False
+
+flowx_cfg = FlowxConfig()
+
+# Config structure for the taskx
+class TaskxConfig:
+    disable_nautic: bool = False
+
+taskx_cfg = TaskxConfig()