-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathpruning.py
269 lines (217 loc) · 9.44 KB
/
pruning.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
import torch
import torch.nn as nn
import numpy as np
import random
import argparse
import pdb
import os
from tqdm import tqdm
import dgl
from dgl.nn.pytorch import GraphConv
import torch.nn.utils.prune as prune
device = torch.device('cuda')
def setup_seed(seed):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
np.random.seed(seed)
random.seed(seed)
def print_args(args, str_num=80):
for arg, val in args.__dict__.items():
print(arg + '.' * (str_num - len(arg) - len(str(val))) + str(val))
print()
def parser_loader():
parser = argparse.ArgumentParser(description='GNN baselines on ogbg-ppa data with Pytorch Geometrics')
parser.add_argument('--pa', type=float, default=0)
parser.add_argument('--pw', type=float, default=0)
parser.add_argument('--irm_lambda', type=float, default=0)
parser.add_argument('--lambda_dis', type=float, default=1.)
parser.add_argument('--lambda_swap', type=float, default=1.)
parser.add_argument('--q', type=float, default=0.7)
parser.add_argument('--masker_lr', type=float, default=0.01)
parser.add_argument('--seed', type=int,help="Please give a value for seed")
parser.add_argument('--config', help="Please give a config.json file with training/model/data/param details")
parser.add_argument('--epochs', type=int, help="Please give a value for epochs")
parser.add_argument('--mask_epochs', type=int, help="Please give a value for epochs")
parser.add_argument('--swap_epochs', type=int, help="Please give a value for epochs")
parser.add_argument('--eval_epochs', type=int, help="Please give a value for epochs")
parser.add_argument('--use_mask', type=int, help="Please give a value for epochs")
parser.add_argument('--gpu_id', help="Please give a value for gpu id")
parser.add_argument('--model', help="Please give a value for model name")
parser.add_argument('--dataset', help="Please give a value for dataset name")
parser.add_argument('--data_dir', help="Please give a value for dataset dir")
parser.add_argument('--out_dir', help="Please give a value for out_dir")
return parser
"""
GPU Setup
"""
def gpu_setup(use_gpu, gpu_id):
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = str(gpu_id)
if torch.cuda.is_available() and use_gpu:
print('cuda available with GPU:',torch.cuda.get_device_name(0))
device = torch.device("cuda")
else:
print('cuda not available')
device = torch.device("cpu")
return device
def prRed(skk): print("\033[91m{}\033[00m".format(skk))
def prGreen(skk): print("\033[92m{}\033[00m".format(skk))
def prYellow(skk): print("\033[93m{}\033[00m".format(skk))
def pruning_batch_data_from_mask(data_list, batch_labels, data_mask, args):
offset = 0
nodes_offset = 0
new_data_list = []
label_list = []
for data, label in zip(data_list, batch_labels):
num_edges = data.number_of_edges()
edge_score = data_mask[offset:offset + num_edges]
mean_score = torch.mean(edge_score)
sorted_score, index = torch.sort(edge_score.view(-1))
prune_num_edges = int(num_edges * 0.8)
#_, index = torch.sort(edge_score.view(-1))
prune_index = index[:prune_num_edges]
data.remove_edges(prune_index)
#data.edata['mask'] = edge_score.cpu()
isolated_nodes = ((data.in_degrees() == 0) & (data.out_degrees() == 0)).nonzero().squeeze(1)
data.remove_nodes(isolated_nodes)
new_data_list.append(data)
label_list.append(label)
offset += num_edges
return new_data_list, label_list
def pruning_batch_data_from_mask_cp(data_list, batch_labels, data_mask, args):
offset = 0
nodes_offset = 0
new_data_list = []
label_list = []
for data, label in zip(data_list, batch_labels):
num_edges = data.number_of_edges()
edge_score = data_mask[offset:offset + num_edges]
mean_score = torch.mean(edge_score)
sorted_score, index = torch.sort(edge_score.view(-1))
prune_num_edges = int(torch.sum(torch.lt(edge_score, mean_score)))
#_, index = torch.sort(edge_score.view(-1))
prune_index = index[:prune_num_edges]
data.remove_edges(prune_index)
#data.edata['mask'] = edge_score.cpu()
isolated_nodes = ((data.in_degrees() == 0) & (data.out_degrees() == 0)).nonzero().squeeze(1)
data.remove_nodes(isolated_nodes)
new_data_list.append(data)
label_list.append(label)
offset += num_edges
return new_data_list, label_list
class DGLFormDataset(torch.utils.data.Dataset):
"""
DGLFormDataset wrapping graph list and label list as per pytorch Dataset.
*lists (list): lists of 'graphs' and 'labels' with same len().
"""
def __init__(self, *lists):
assert all(len(lists[0]) == len(li) for li in lists)
self.lists = lists
self.graph_lists = lists[0]
self.graph_labels = lists[1]
def __getitem__(self, index):
return tuple(li[index] for li in self.lists)
def __len__(self):
return len(self.lists[0])
def masker_pruning_dataset(dataloader, masker, device, args):
data_list = []
label_list = []
with torch.no_grad():
for iter, (batch_graphs, batch_labels) in enumerate(dataloader):
batch_graphs = dgl.batch(batch_graphs).to(device)
batch_x = batch_graphs.ndata['feat'].to(device) # num x feat
#batch_e = batch_graphs.edata['feat'].to(device)
batch_e = None
data_mask, node_masker = masker(batch_graphs, batch_x, batch_e)
batch_graphs = dgl.unbatch(batch_graphs)
batch_graphs_pruned, label_pruned = pruning_batch_data_from_mask(batch_graphs, batch_labels, data_mask, args)
data_list += batch_graphs_pruned
label_list += label_pruned
#pruned_data = DGLFormDataset(data_list, label_list)
return data_list, label_list
def see_zero_rate(model):
sum_list = 0
zero_sum = 0
for m in model.modules():
if isinstance(m, nn.Linear):
sum_list = sum_list + float(m.weight.nelement())
zero_sum = zero_sum + float(torch.sum(m.weight == 0))
print('INFO: Weight Sparsity [{:.4f}%] '.format(100 * (zero_sum / sum_list)))
return zero_sum / sum_list
def extract_mask(model):
model_dict = model.state_dict()
new_dict = {}
for key in model_dict.keys():
if 'mask' in key:
new_dict[key] = model_dict[key]
return new_dict
def pruning_model_by_mask(model, mask_dict):
module_to_prune = []
mask_to_prune = []
module_to_prune.append(model.embedding_h)
mask_to_prune.append(mask_dict['embedding_h.weight_mask'])
module_to_prune.append(model.layers[0].apply_mod.linear)
mask_to_prune.append(mask_dict['layers.0.apply_mod.linear.weight_mask'])
module_to_prune.append(model.layers[1].apply_mod.linear)
mask_to_prune.append(mask_dict['layers.1.apply_mod.linear.weight_mask'])
#module_to_prune.append(model.layers[2].apply_mod.linear)
#mask_to_prune.append(mask_dict['layers.2.apply_mod.linear.weight_mask'])
#module_to_prune.append(model.layers[3].apply_mod.linear)
#mask_to_prune.append(mask_dict['layers.3.apply_mod.linear.weight_mask'])
#module_to_prune.append(model.MLP_layer.FC_layers[0])
#mask_to_prune.append(mask_dict['MLP_layer.FC_layers.0.weight_mask'])
#module_to_prune.append(model.MLP_layer.FC_layers[1])
#mask_to_prune.append(mask_dict['MLP_layer.FC_layers.1.weight_mask'])
print("linear", model.MLP_layer)
module_to_prune.append(model.MLP_layer)
mask_to_prune.append(mask_dict['MLP_layer.weight_mask'])
for ii in range(len(module_to_prune)):
prune.CustomFromMask.apply(module_to_prune[ii], 'weight', mask=mask_to_prune[ii])
def pruning_model(model, px, random=False):
if px == 0:
pass
else:
parameters_to_prune =[]
for m in model.modules():
if isinstance(m, nn.Linear):
parameters_to_prune.append((m,'weight'))
print(m)
parameters_to_prune = tuple(parameters_to_prune)
if random:
prune.global_unstructured(
parameters_to_prune,
pruning_method=prune.RandomUnstructured,
amount=px,
)
else:
prune.global_unstructured(
parameters_to_prune,
pruning_method=prune.L1Unstructured,
amount=px,
)
def print_pruning_percent(dataset_ori, dataset_pru, str1):
ori_all = 0.0
pru_all = 0.0
for data_ori, data_pru in zip(dataset_ori, dataset_pru):
#print(str1, data_ori[1], data_ori[2])
ori = dgl.batch(data_ori[0]).number_of_edges()
pru = dgl.batch(data_pru[0]).number_of_edges()
ori_all += ori
pru_all += pru
sp = 1 - pru_all / ori_all
# print('INFO: Dataset Sparsity [{:.4f}%] '.format(100 * sp))
return sp
def plot_mask(data_mask):
a = (data_mask <= 0.2).sum()
b = (data_mask <= 0.4).sum()
c = (data_mask <= 0.6).sum()
d = (data_mask <= 0.8).sum()
e = (data_mask <= 1.0).sum()
a, b, c, d, e = float(a), float(b), float(c), float(d), float(e)
a1 = a / e # (0.0 - 0.2)
b1 = (b - a) / e # (0.2 - 0.4)
c1 = (c - b) / e # (0.4 - 0.6)
d1 = (d - c) / e # (0.6 - 0.8)
e1 = (e - d) / e # (0.8 - 1.0)
return [a1, b1, c1, d1, e1]