Refactored hog + small neural network classification pipeline.

.vscode/launch.json

@@ -418,6 +418,14 @@
                     "--aug", "fa_reduced_cifar10"
             ]
         },
+        {
+            "name": "Train Visual Features FF Net",
+            "type": "python",
+            "request": "launch",
+            "program": "${cwd}/scripts/misc/train_visual_features_ffnet.py",
+            "console": "integratedTerminal",
+            "args": []
+        },
         {
             "name": "Analysis Aggregate",
             "type": "python",

archai/algos/img_classif_nondl/linear_model_evaluator.py

@@ -1,55 +0,0 @@
-# Copyright (c) Microsoft Corporation.
-# Licensed under the MIT license.
-
-from copy import deepcopy
-from typing import Optional
-import importlib
-import sys
-import string
-import os
-
-from overrides import overrides
-
-import torch
-from torch import nn
-
-from overrides import overrides, EnforceOverrides
-
-from archai.common.trainer import Trainer
-from archai.common.config import Config
-from archai.common.common import logger
-from archai.datasets import data
-from archai.nas.model_desc import ModelDesc
-from archai.nas.model_desc_builder import ModelDescBuilder
-from archai.nas import nas_utils
-from archai.common import ml_utils, utils
-from archai.common.metrics import EpochMetrics, Metrics
-from archai.nas.model import Model
-from archai.common.checkpoint import CheckPoint
-from archai.nas.evaluater import Evaluater
-from archai.algos.proxynas.freeze_trainer import FreezeTrainer
-from archai.algos.proxynas.conditional_trainer import ConditionalTrainer
-
-from nats_bench import create
-from archai.algos.natsbench.lib.models import get_cell_based_tiny_net
-
-class LinearEvaluater(Evaluater):
-    @overrides
-    def create_model(self, conf_eval:Config, model_desc_builder:ModelDescBuilder,
-                      final_desc_filename=None, full_desc_filename=None)->nn.Module:
-        # region conf vars
-        dataset_name = conf_eval['loader']['dataset']['name']
-
-        # if explicitly passed in then don't get from conf
-        if not final_desc_filename:
-            final_desc_filename = conf_eval['final_desc_filename']
-
-        dataroot = utils.full_path(conf_eval['loader']['dataset']['dataroot'])    
-        # endregion
-
-        # create linear model
-
-        # problem: what is the input size of the image features?
-
-
-        return self._model_from_natsbench(arch_index, dataset_name, natsbench_location)

archai/networks/visual_features_with_ff_net.py

@@ -1,17 +1,22 @@
 import torch
 from torch.functional import Tensor
 import torch.nn as nn
+
 from skimage import feature
-import matplotlib.pyplot as plt
 
 
-
-class VisualFeaturesWithLinearNet(nn.Module):
+class VisualFeaturesWithFFNet(nn.Module):
     def __init__(self, feature_len:int, n_classes:int):
-        super(VisualFeaturesWithLinearNet, self).__init__()
+        super(VisualFeaturesWithFFNet, self).__init__()
         self.feature_len = feature_len
 
-        self.fc = nn.Linear(feature_len, n_classes)
+        self.net = nn.Sequential(nn.Linear(feature_len, 2048),
+                                nn.ReLU(),
+                                nn.Linear(2048, 2048),
+                                nn.ReLU(),
+                                nn.Linear(2048, n_classes))
+        
+        #self.fc = nn.Linear(feature_len, n_classes)
 
 
     def _compute_features(self, x:Tensor)->Tensor:
@@ -41,4 +46,4 @@ class VisualFeaturesWithLinearNet(nn.Module):
 
     def forward(self, x):
         feats = self._compute_features(x)    
-        return self.fc(feats.float())
+        return self.net(feats.float())

confs/algos/visual_features_ffnet.yaml

@@ -1,5 +1,12 @@
 __include__: 'darts.yaml' # just use darts defaults
 
+dataset:
+  name: 'ImageNet16-120'
+  n_classes: 120
+  channels: 3 # number of channels in image
+  max_batches: -1 # if >= 0 then only these many batches are generated (useful for debugging)
+  storage_name: 'imagenet16' # name of folder or tar file to copy from cloud storage
+
 common:
   experiment_name: 'VisualFeaturesWithLinearNet'
 

scripts/misc/train_visual_features_ffnet.py

@@ -2,7 +2,7 @@
 # Licensed under the MIT license.
 
 import torch
-from archai.networks.visual_features_with_linear_net import VisualFeaturesWithLinearNet
+from archai.networks.visual_features_with_ff_net import VisualFeaturesWithFFNet
 from archai.common.trainer import Trainer
 from archai.common.config import Config
 from archai.common.common import common_init
@@ -13,7 +13,7 @@ def train_test(conf_eval:Config):
     conf_trainer = conf_eval['trainer']
 
     # create model
-    Net = VisualFeaturesWithLinearNet
+    Net = VisualFeaturesWithFFNet
     feature_len = 324
     n_classes = 10
     model = Net(feature_len, n_classes).to(torch.device('cuda',  0))
@@ -27,7 +27,7 @@ def train_test(conf_eval:Config):
 
 
 if __name__ == '__main__':
-    conf = common_init(config_filepath='confs/algos/visual_features_linearnet.yaml')
+    conf = common_init(config_filepath='confs/algos/visual_features_ffnet.yaml')
     conf_eval = conf['nas']['eval']
 
     train_test(conf_eval)

scripts/reports/proxynas_plots/cross_exp_conf.yaml

@@ -32,6 +32,8 @@ natsbench_cifar10:
                 'ft_fb256_ftlr0.1_fte5_ct256_ftt0.6_c9',
                 'ft_fb256_ftlr0.1_fte10_ct256_ftt0.6_c9']
 
+  zero_cost: ['zero_cost_nb201_cifar10']
+
   shortreg: ['nb_reg_b1024_e01', 
             'nb_reg_b1024_e02', 
             'nb_reg_b1024_e04', 
@@ -67,6 +69,8 @@ natsbench_cifar100:
             'nb_c100_reg_b2048_e20', 
             'nb_c100_reg_b2048_e30']
 
+  zero_cost: ['zero_cost_nb201_cifar100']
+
 natsbench_imagenet16-120:
   freezetrain: [ft_i6_fb2048_ftlr0.1_fte5_ct256_ftt0.1, 
                 ft_i6_fb2048_ftlr0.1_fte10_ct256_ftt0.1,
@@ -84,6 +88,8 @@ natsbench_imagenet16-120:
                 ft_i6_fb512_ftlr0.1_fte10_ct256_ftt0.2,
                 ft_i6_fb256_ftlr0.1_fte5_ct256_ftt0.2]
 
+  zero_cost: ['zero_cost_nb201_ImageNet16-120']
+
   shortreg: [nb_i16_reg_b256_e10,
             nb_i16_reg_b256_e20,
             nb_i16_reg_b256_e30,

scripts/reports/proxynas_plots/cross_exp_plots.py

@@ -14,6 +14,8 @@ import plotly.express as px
 from plotly.subplots import make_subplots
 import plotly.graph_objects as go
 
+ZERO_COST_MEASURES = ['fisher', 'grad_norm', 'grasp', 'jacob_cov', 'plain', 'snip', 'synflow', 'synflow_bn']
+
 
 def parse_raw_data(root_exp_folder:str, exp_list:List[str])->Dict:
     data = {}
@@ -26,7 +28,7 @@ def parse_raw_data(root_exp_folder:str, exp_list:List[str])->Dict:
 
 def main():
     parser = argparse.ArgumentParser(description='Cross Experiment Plots')
-    parser.add_argument('--dataset', type=str, default='nasbench101',
+    parser.add_argument('--dataset', type=str, default='natsbench_imagenet16-120',
                         help='dataset on which experiments have been run')
     parser.add_argument('--conf_location', type=str, default='scripts/reports/proxynas_plots/cross_exp_conf.yaml', 
                         help='location of conf file')
@@ -39,10 +41,12 @@ def main():
 
     exp_list = conf_data[args.dataset]['freezetrain']
     shortreg_exp_list = conf_data[args.dataset]['shortreg']
+    zero_cost_exp_list = conf_data[args.dataset]['zero_cost']
                         
     # parse raw data from all processed experiments
     data = parse_raw_data(exp_folder, exp_list)
     shortreg_data = parse_raw_data(exp_folder, shortreg_exp_list)
+    zero_cost_data = parse_raw_data(exp_folder, zero_cost_exp_list)
 
     # collect linestyles and colors to create distinguishable plots
     cmap = plt.get_cmap('tab20')
@@ -126,6 +130,15 @@ def main():
             if j == 0:
                 break
 
+        # get zero cost measures
+        for j, key in enumerate(zero_cost_data.keys()):
+            assert tp == zero_cost_data[key]['top_percents'][i]
+            for measure in ZERO_COST_MEASURES:
+                spe_name = measure + '_spe'
+                cr_name = measure + '_ratio_common'
+                this_tp_info['zero_cost_' + measure] = (0.0, zero_cost_data[key][spe_name][i], zero_cost_data[key][cr_name][i])
+
+
         # get shortreg
         for key in shortreg_data.keys():
             exp_name = key
@@ -145,6 +158,7 @@ def main():
     for ind, tp_key in enumerate(tp_info.keys()):
         counter = 0
         counter_reg = 0
+        counter_zero = 0
         for exp in tp_info[tp_key].keys():
             duration = tp_info[tp_key][exp][0]
             spe = tp_info[tp_key][exp][1]
@@ -158,6 +172,10 @@ def main():
                 marker = counter_reg
                 marker_color = 'blue'
                 counter_reg += 1
+            elif 'zero_cost' in exp:
+                marker = counter_zero
+                marker_color = 'green'
+                counter_zero += 1            
             else:
                 raise NotImplementedError