blob: 8664a1512ae7a1f93dd4107cd52794150cd2ab7a (
plain)
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
|
experiment_group: Sample Experiments
experiments:
- train_args:
batch_size: 256
max_epochs: &max_epochs 32
dataset:
type: EmnistDataset
args:
sample_to_balance: true
subsample_fraction: null
transform: null
target_transform: null
seed: 4711
train_args:
num_workers: 6
train_fraction: 0.8
model: CharacterModel
metrics: [accuracy]
# network: MLP
# network_args:
# input_size: 784
# hidden_size: 512
# output_size: 80
# num_layers: 5
# dropout_rate: 0.2
# activation_fn: SELU
network:
type: ResidualNetwork
args:
in_channels: 1
num_classes: 80
depths: [2, 2]
block_sizes: [64, 64]
activation: leaky_relu
stn: true
# network:
# type: WideResidualNetwork
# args:
# in_channels: 1
# num_classes: 80
# depth: 10
# num_layers: 3
# width_factor: 4
# dropout_rate: 0.2
# activation: SELU
# network: LeNet
# network_args:
# output_size: 62
# activation_fn: GELU
criterion:
type: CrossEntropyLoss
args:
weight: null
ignore_index: -100
reduction: mean
optimizer:
type: AdamW
args:
lr: 1.e-02
betas: [0.9, 0.999]
eps: 1.e-08
# weight_decay: 5.e-4
amsgrad: false
# lr_scheduler:
# type: OneCycleLR
# args:
# max_lr: 1.e-03
# epochs: *max_epochs
# anneal_strategy: linear
lr_scheduler:
type: CosineAnnealingLR
args:
T_max: *max_epochs
interval: epoch
swa_args:
start: 2
lr: 5.e-2
callbacks: [Checkpoint, ProgressBar, WandbCallback, WandbImageLogger, EarlyStopping]
callback_args:
Checkpoint:
monitor: val_accuracy
ProgressBar:
epochs: null
log_batch_frequency: 100
EarlyStopping:
monitor: val_loss
min_delta: 0.0
patience: 5
mode: min
WandbCallback:
log_batch_frequency: 10
WandbImageLogger:
num_examples: 4
use_transpose: true
verbosity: 0 # 0, 1, 2
resume_experiment: null
test: true
test_metric: test_accuracy
|
