-
Notifications
You must be signed in to change notification settings - Fork 1
/
main.py
156 lines (122 loc) · 4.37 KB
/
main.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
import os
import sys
import csv
import numpy as np
import torch
import math
import torch.nn as nn
import torch.optim as optim
from torch.utils import data
from sourcerer import SourceRegLoss
from cnn_model import ConvModel
from sits_data import SITSData
def main():
source_tile = sys.argv[1]
target_tile = sys.argv[2]
run_no = sys.argv[3]
polygons = sys.argv[4]
main_path = "/home/bluc0001/nc23_scratch/ben/"
exp_name = "Sourcerer_s_"+source_tile+"_t_"+target_tile
NO_CLASSES = 30
data_path = "/home/bluc0001/nc23_scratch/ben/data/"
NO_BANDS = 10
results_path = main_path+"Results/"
seed = 18
np.random.seed(seed)
torch.manual_seed(seed)
##### INITIALISE MODEL #####
cnn = ConvModel()
##### LOSS FUNCTION AND OPTIMIZER #####
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(cnn.parameters())
model_filepath = results_path+exp_name+"_run_"+run_no+"_model.pth"
if os.path.isfile(model_filepath):
print("Model Found, skipping training on Source")
cnn = torch.load(model_filepath)
else:
print("Training on Source Train data...")
##### LOADING SOURCE TRAINING DATA #####
source_train_data_file = data_path+source_tile+"/"+source_tile+"_train.csv"
source_train = SITSData(source_train_data_file, n_channels=NO_BANDS)
source_train_generator = data.DataLoader(source_train, batch_size=32, drop_last=True)
##### TRAIN MODEL #####
cnn.train()
correct_val = 0
total_val = 0
loss_list = []
for i, (X, y) in enumerate(source_train_generator):
# Run the forward pass
predictions = cnn(X.float())
loss = criterion(predictions, y)
loss_list.append(loss.item())
# Backprop and optimise
optimizer.zero_grad()
loss.backward()
optimizer.step()
# Track the accuracy
total_val += y.size(0)
_, predicted = torch.max(predictions.data, dim=1)
correct_val += (predicted == y).sum().item()
train_accuracy = correct_val/total_val
print("Train Accuracy: ", train_accuracy)
del source_train, source_train_generator
torch.save(cnn, model_filepath)
print("Training on "+str(polygons)+" Target Polygons")
if int(polygons)==0:
target_train_qty = 0
epochs_required = 0
reg_constant = 0
else:
tfr_model_filename = results_path+exp_name+"_run_"+run_no+"_pgns_"+str(polygons)+"_model.pth"
if os.path.isfile(tfr_model_filename):
print("Model found... exiting...")
sys.exit()
for module in cnn.modules():
if isinstance(module, nn.BatchNorm1d):
module.eval()
optimizer = optim.Adam(cnn.parameters())
#### LOADING TARGET TRAIN DATA #####
target_train_data_file = data_path+target_tile+"/"+target_tile+"_subsets_run_poly/"+"run_"+run_no+"_polygons_"+str(polygons)+".csv"
target_train_qty = sum(1 for line in open(target_train_data_file))-1
print("Target train qty: ", target_train_qty)
target_train = SITSData(target_train_data_file, n_channels=NO_BANDS)
batch_sz = 32
print("Batch Size: ", batch_sz)
target_train_generator = data.DataLoader(target_train, batch_size = batch_sz)
source_reg_loss = SourceRegLoss(cnn, target_train_qty)
no_updates = 5000
print("No updates: ", no_updates)
epochs_required = math.ceil(no_updates * batch_sz / target_train_qty)
print("No epochs: ", epochs_required)
for epoch in range(epochs_required):
for i, (X, y) in enumerate(target_train_generator, 0):
# Run the forward pass
predictions = cnn(X.float())
loss = source_reg_loss(predictions, y, cnn)
print("Loss: ", loss.data)
# Backprop and optimise
optimizer.zero_grad()
loss.backward()
optimizer.step()
torch.save(cnn, tfr_model_filename)
del target_train, target_train_generator
#### LOADING TARGET TEST DATA #####
target_test_data_file = data_path+target_tile+"/"+target_tile+"_test.csv"
target_test = SITSData(target_test_data_file, n_channels=NO_BANDS)
target_test_generator = data.DataLoader(target_test, batch_size=10_000)
##### PREDICT TEST DATA #####
cnn.eval()
with torch.no_grad():
correct_test = 0
total_test = 0
for i, (X, y) in enumerate(target_test_generator):
predictions = cnn(X.float())
_, predicted = torch.max(predictions.data, dim=1)
total_test += y.size(0)
correct_test += (predicted == y).sum().item()
test_accuracy = correct_test/total_test
print("--------------------------------------------------------")
print("Total test acc: ", test_accuracy)
print("--------------------------------------------------------")
if __name__ == "__main__":
main()