/
locationencoder.py
160 lines (133 loc) · 6.07 KB
/
locationencoder.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
from torch import optim, nn
import lightning.pytorch as pl
import locationencoder.pe as PE
import locationencoder.nn as NN
from utils.losses import get_loss_fn
from sklearn.metrics import (
accuracy_score,
jaccard_score,
mean_absolute_error
)
def get_positional_encoding(name, hparams=None):
if name == "direct":
return PE.Direct()
elif name == "cartesian3d":
return PE.Cartesian3D()
elif name == "sphericalharmonics":
# default to analytic
if "harmonics_calculation" not in hparams.keys():
hparams["harmonics_calculation"] = "analytic"
if "harmonics_calculation" in hparams.keys() and hparams['harmonics_calculation'] == "discretized":
return PE.DiscretizedSphericalHarmonics(legendre_polys=hparams['legendre_polys'])
else:
return PE.SphericalHarmonics(legendre_polys=hparams['legendre_polys'],
harmonics_calculation=hparams['harmonics_calculation'])
elif name == "theory":
return PE.Theory(min_radius=hparams['min_radius'],
max_radius=hparams['max_radius'],
frequency_num=hparams['frequency_num'])
elif name == "wrap":
return PE.Wrap()
elif name in ["grid", "spherec", "spherecplus", "spherem", "spheremplus"]:
return PE.GridAndSphere(min_radius=hparams['min_radius'],
max_radius=hparams['max_radius'],
frequency_num=hparams['frequency_num'],
name=name)
else:
raise ValueError(f"{name} not a known positional encoding.")
def get_neural_network(name, input_dim, hparams=None):
if name == "linear":
return nn.Linear(input_dim, hparams['num_classes'])
elif name == "siren":
return NN.SirenNet(
dim_in=input_dim,
dim_hidden=hparams['dim_hidden'],
num_layers=hparams['num_layers'],
dim_out=hparams['num_classes'],
dropout=hparams['dropout'] if "dropout" in hparams.keys() else False
)
elif name == "fcnet":
return NN.FCNet(
num_inputs=input_dim,
num_classes=hparams['num_classes'],
dim_hidden=hparams['dim_hidden']
)
else:
raise ValueError(f"{name} not a known neural networks.")
def get_param(hparams, key, default=False):
"""
Convenience function that indexes the hyperparameter dict but returns a default value if not defined rather than
an error
"""
return hparams[key] if key in hparams.keys() else default
# define the LightningModule
class LocationEncoder(pl.LightningModule):
def __init__(self, positional_encoding_name, neural_network_name, hparams):
super().__init__()
self.learning_rate = hparams["optimizer"]["lr"]
self.weight_decay = hparams["optimizer"]["wd"]
self.regression = get_param(hparams, "regression")
self.loss_fn = get_loss_fn(presence_only=get_param(hparams, "presence_only_loss"),
loss_weight=get_param(hparams, "loss_weight"),
regression=self.regression)
self.positional_encoder = get_positional_encoding(
positional_encoding_name, hparams
)
self.neural_network = get_neural_network(
neural_network_name,
input_dim=self.positional_encoder.embedding_dim,
hparams=hparams
)
# this enables LocationEncoder.load_from_checkpoint(path)
self.save_hyperparameters()
def common_step(self, batch, batch_idx):
lonlats, label = batch
return self.loss_fn(self, lonlats, label)
def forward(self, lonlats):
embedding = self.positional_encoder(lonlats)
return self.neural_network(embedding)
def training_step(self, batch, batch_idx):
loss = self.common_step(batch, batch_idx)
return loss
def validation_step(self, batch, batch_idx):
loss = self.common_step(batch, batch_idx)
self.log("val_loss", loss, on_step=False, on_epoch=True)
return {"val_loss":loss}
def predict_step(self, batch, batch_idx):
lonlats, label = batch
prediction_logits = self.forward(lonlats)
return prediction_logits, lonlats, label
def test_step(self, batch, batch_idx):
lonlats, label = batch
prediction_logits = self.forward(lonlats)
loss = self.loss_fn(self, lonlats, label)
# check if binary
non_binary_task = self.regression
if (prediction_logits.size(1) == 1) and not (non_binary_task):
y_pred = (prediction_logits.squeeze() > 0).cpu()
average = "binary"
elif self.regression:
y_pred = prediction_logits.cpu()
else: # take argmax
y_pred = prediction_logits.argmax(-1).cpu()
average = "macro"
self.log("test_loss", loss, on_step=False, on_epoch=True)
if self.regression:
MAE = mean_absolute_error(y_true=label.cpu(), y_pred = y_pred)
self.log("test_MAE", MAE, on_step=False, on_epoch=True)
test_results = {"test_loss":loss,
"test_MAE":MAE}
else:
accuracy = accuracy_score(y_true=label.cpu(), y_pred= y_pred).astype("float32")
IoU = jaccard_score(y_true=label.cpu(), y_pred = y_pred, average=average).astype("float32")
self.log("test_accuracy", accuracy, on_step=False, on_epoch=True)
self.log("test_IoU", IoU, on_step=False, on_epoch=True)
test_results = {"test_loss":loss,
"test_accuracy":accuracy}
return test_results
def configure_optimizers(self):
optimizer = optim.Adam([{"params": self.neural_network.parameters()},
{"params": self.positional_encoder.parameters(), "weight_decay":0}],
lr=self.learning_rate,
weight_decay=self.weight_decay)
return optimizer