generated from kyegomez/Paper-Implementation-Template
-
-
Notifications
You must be signed in to change notification settings - Fork 0
/
tests.py
335 lines (263 loc) · 11.8 KB
/
tests.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
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
import pytest
import torch
from torch import nn
from torch.optim import SGD
from ge.main import GradientEquilibrum
# Helper function to create a simple model and loss for testing
def create_model_and_loss():
dim_in = 2
dim_out = 1
model = torch.nn.Linear(dim_in, dim_out)
loss_fn = torch.nn.MSELoss()
return model, loss_fn
# Test optimizer with default parameters
def test_optimizer_default_parameters():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters())
assert isinstance(optimizer, GradientEquilibrum)
assert optimizer.defaults["lr"] == 0.01
assert optimizer.defaults["max_iterations"] == 1000
assert optimizer.defaults["tol"] == 1e-7
assert optimizer.defaults["weight_decay"] == 0.0
# Test optimizer step function with zero gradient
def test_optimizer_step_with_zero_gradient():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters())
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[0.0, 0.0]]), torch.tensor([[0.0]])))
loss.backward()
optimizer.step()
assert True # No exceptions were raised
# Test optimizer step function with a non-zero gradient
def test_optimizer_step_with_non_zero_gradient():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters())
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step()
assert True # No exceptions were raised
# Test optimizer step function with weight decay
def test_optimizer_step_with_weight_decay():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), weight_decay=0.1)
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step()
assert True # No exceptions were raised
# Test optimizer clip_grad_value function
def test_optimizer_clip_grad_value():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters())
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.clip_grad_value(0.1)
optimizer.step()
assert True # No exceptions were raised
# Test optimizer add_weight_decay function
def test_optimizer_add_weight_decay():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters())
optimizer.add_weight_decay(0.1)
assert optimizer.param_groups[0]["weight_decay"] == 0.1
# Test optimizer state_dict and load_state_dict functions
def test_optimizer_state_dict_and_load_state_dict():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters())
state_dict = optimizer.state_dict()
optimizer.load_state_dict(state_dict)
assert optimizer.defaults == state_dict["param_groups"][0]
assert optimizer.state == state_dict["state"]
# Test optimizer with a custom learning rate
def test_optimizer_with_custom_lr():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), lr=0.1)
assert optimizer.defaults["lr"] == 0.1
# Test optimizer with a custom max_iterations
def test_optimizer_with_custom_max_iterations():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), max_iterations=500)
assert optimizer.defaults["max_iterations"] == 500
# Test optimizer with a custom tolerance
def test_optimizer_with_custom_tolerance():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), tol=1e-6)
assert optimizer.defaults["tol"] == 1e-6
# Test optimizer with a custom learning rate and weight decay
def test_optimizer_with_custom_lr_and_weight_decay():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), lr=0.1, weight_decay=0.2)
assert optimizer.defaults["lr"] == 0.1
assert optimizer.defaults["weight_decay"] == 0.2
# Test optimizer with a custom clip threshold
def test_optimizer_with_custom_clip_threshold():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), clip_thresh=0.5)
assert True # No exceptions were raised
# Test optimizer with custom parameters and custom learning rate
def test_optimizer_with_custom_parameters_and_lr():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(
model.parameters(), lr=0.1, max_iterations=500, tol=1e-6, weight_decay=0.2
)
assert optimizer.defaults["lr"] == 0.1
assert optimizer.defaults["max_iterations"] == 500
assert optimizer.defaults["tol"] == 1e-6
assert optimizer.defaults["weight_decay"] == 0.2
# Test optimizer with a large learning rate and max_iterations
def test_optimizer_with_large_lr_and_max_iterations():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), lr=1e3, max_iterations=10000)
assert optimizer.defaults["lr"] == 1e3
assert optimizer.defaults["max_iterations"] == 10000
# Test optimizer with a very small tolerance
def test_optimizer_with_small_tolerance():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), tol=1e-10)
assert optimizer.defaults["tol"] == 1e-10
# Test optimizer step function with a custom closure
def test_optimizer_step_with_custom_closure():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters())
# Custom closure that computes and returns loss
def custom_closure():
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
return loss
loss = optimizer.step(closure=custom_closure)
assert isinstance(loss, torch.Tensor)
# Test optimizer with custom parameters and weight decay
def test_optimizer_with_custom_parameters_and_weight_decay():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(
model.parameters(),
lr=0.1,
max_iterations=500,
tol=1e-6,
weight_decay=0.2,
)
assert optimizer.defaults["lr"] == 0.1
assert optimizer.defaults["max_iterations"] == 500
assert optimizer.defaults["tol"] == 1e-6
assert optimizer.defaults["weight_decay"] == 0.2
# Test optimizer step function with custom learning rate
def test_optimizer_step_with_custom_lr():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), lr=0.1)
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step(lr=0.01) # Custom learning rate for this step
assert True # No exceptions were raised
# Test optimizer step function with a very small learning rate
def test_optimizer_step_with_small_lr():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), lr=0.1)
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step(lr=1e-6) # Very small learning rate for this step
assert True # No exceptions were raised
# Test optimizer step function with a custom clip threshold
def test_optimizer_step_with_custom_clip_threshold():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), clip_thresh=0.5)
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step()
assert True # No exceptions were raised
# Test optimizer step function with weight decay and custom learning rate
def test_optimizer_step_with_weight_decay_and_custom_lr():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), lr=0.1, weight_decay=0.2)
optimizer.zero_grad()
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step(lr=0.01) # Custom learning rate for this step
assert True # No exceptions were raised
# Test optimizer step function with custom gradient values
def test_optimizer_step_with_custom_gradient_values():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters())
optimizer.zero_grad()
# Custom gradients for testing
custom_gradients = [torch.tensor([[-1.0, -1.0]])]
for param, grad in zip(model.parameters(), custom_gradients):
param.grad = grad
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step()
# Check if the parameters were updated correctly
for param, grad in zip(model.parameters(), custom_gradients):
assert torch.allclose(param.data, grad, atol=1e-7)
# Test optimizer step function with custom gradient values and clip threshold
def test_optimizer_step_with_custom_gradient_values_and_clip_threshold():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), clip_thresh=0.5)
optimizer.zero_grad()
# Custom gradients for testing
custom_gradients = [torch.tensor([[-1.0, -1.0]])]
for param, grad in zip(model.parameters(), custom_gradients):
param.grad = grad
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step()
# Check if the parameters were updated correctly and clipped
for param, grad in zip(model.parameters(), custom_gradients):
clipped_grad = torch.clamp(grad, -0.5, 0.5)
assert torch.allclose(param.data, clipped_grad, atol=1e-7)
# Test optimizer step function with custom gradient values and weight decay
def test_optimizer_step_with_custom_gradient_values_and_weight_decay():
model, loss_fn = create_model_and_loss()
optimizer = GradientEquilibrum(model.parameters(), weight_decay=0.1)
optimizer.zero_grad()
# Custom gradients for testing
custom_gradients = [torch.tensor([[-1.0, -1.0]])]
for param, grad in zip(model.parameters(), custom_gradients):
param.grad = grad
loss = loss_fn(model(torch.tensor([[1.0, 1.0]]), torch.tensor([[1.0]])))
loss.backward()
optimizer.step()
# Check if the parameters were updated correctly with weight decay
for param, grad in zip(model.parameters(), custom_gradients):
updated_param = grad - 0.1 * grad
assert torch.allclose(param.data, updated_param, atol=1e-7)
# Define a sample model and data
class SampleModel(nn.Module):
def __init__(self):
super(SampleModel, self).__init__()
self.fc = nn.Linear(10, 10)
def forward(self, x):
return self.fc(x)
# Define a benchmark function
@pytest.mark.benchmark(group="optimizer_comparison")
def test_optimizer_performance(benchmark):
# Create a sample model and data
model = SampleModel()
data = torch.randn(64, 10)
target = torch.randn(64, 10)
loss_fn = nn.MSELoss()
# Create instances of your optimizer and an alternative optimizer
custom_optimizer = GradientEquilibrum(model.parameters(), lr=0.01)
sgd_optimizer = SGD(model.parameters(), lr=0.01)
# Benchmark your optimizer's step method
def custom_step():
custom_optimizer.zero_grad()
loss = loss_fn(model(data), target)
loss.backward()
custom_optimizer.step()
# Benchmark the alternative optimizer's step method
def sgd_step():
sgd_optimizer.zero_grad()
loss = loss_fn(model(data), target)
loss.backward()
sgd_optimizer.step()
# Measure and compare execution times
custom_time = benchmark(custom_step)
sgd_time = benchmark(sgd_step)
# Assert that your optimizer is as fast or faster than the alternative
assert custom_time < sgd_time