-
Notifications
You must be signed in to change notification settings - Fork 0
/
report.txt
96 lines (57 loc) · 1.75 KB
/
report.txt
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
Q1.1a
report its performance on the training, validation and test sets
training: 0.4654
validation: 0.4610
test sets: 0.3422
Q1.1b
compare (based on the plots of the train and validation accuracies) the models
obtained using two different learning rates of η = 0.01 and η = 0.001
learning rate = 0.01
training set accuracy: 0.6609
validation set accuracy: 0.6568
final test accuracy: 0.5784
learning rate = 0.001
training set accuracy: 0.6625
validation set accuracy: 0.6639
final test accuracy: 0.5936
Q1.2a
Comment the following claim: “A logistic regression model using pixel values
as features is not as expressive as a multi-layer perceptron using relu activations.
However, training a logistic regression model is easier because it is a convex optimization
problem.” Is this true of false? Justify.
Q1.2b
final test accuracy: 0.7524
loss: 947991.7287 | train acc: 0.7975 | val acc: 0.7865
Q2.1
learning rate = 0.01
final accuracy on the test set:
learning rate = 0.001
final accuracy on the test set:
learning rate = 0.0001
final accuracy on the test set:
Q2.2a
report the best test accuracy and
comment on the differences in both
performance and time of execution
batch size = 16
test accuracy:
batch size = 1024
test accuracy:
Q2.2b
report the best test accuracy
and comment on the differences in performance.
learning rate = 1
test accuracy:
learning rate = 0.1
test accuracy:
learning rate = 0.01
test accuracy:
learning rate = 0.001
test accuracy:
Q2.2c
report the best test accuracy and comment on the differences of
both techniques
default model w/ batch size=256 && epochs=150:
is there overfitting?
default model w/ batch size=256 && epochs=150 && L2 reg=0.0001:
default model w/ batch size=256 && epochs=150 && dropout=0.2: