-
Notifications
You must be signed in to change notification settings - Fork 0
/
preprocess.py
149 lines (114 loc) · 5.05 KB
/
preprocess.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
import math
import numpy as np
from sklearn.model_selection import train_test_split
import torch
from torch.utils.data import TensorDataset, DataLoader
import xarray as xr
selection = {
'variables': [
'geopotential',
'10m_u_component_of_wind',
'10m_v_component_of_wind',
'vertical_velocity',
'10m_wind_speed',
'total_precipitation_6hr',
'total_cloud_cover',
'2m_temperature',
'specific_humidity',
'surface_pressure',
'toa_incident_solar_radiation',
'total_column_water_vapour'
],
"levels": [500],
"time_slice": slice('2016-01-01', '2017-01-01'),
"lat_slice": slice(30,50),
"long_slice": slice(70,90),
}
def preprocess_data(split_percentage=0.8, batch_size=20, use_level=False, window_size=10):
obs_path = 'gs://weatherbench2/datasets/era5/1959-2022-6h-64x32_equiangular_conservative.zarr'
data = xr.open_zarr(obs_path)
print("preprocessing")
print("dataset shape: ", data.sizes)
data = data[selection['variables']].sel(level=selection['levels'], time=selection['time_slice'],
latitude=selection['lat_slice'], longitude=selection['long_slice'])
time_size = data.sizes['time']
level_size = data.sizes['level']
lon_size = data.sizes['longitude']
lat_size = data.sizes['latitude']
feature_size = len(selection['variables'])
dataset_shape = (time_size, level_size, lon_size, lat_size) if use_level else (time_size, lon_size, lat_size)
time_arrays = []
for i, var_name in enumerate(data.data_vars):
var_data = data[var_name].values
num_dims = len(var_data.shape)
time_array = np.empty(dataset_shape, dtype=var_data.dtype)
has_level_size = num_dims == 4
if use_level:
if has_level_size:
time_array[:] = var_data
else:
time_array[:] = np.expand_dims(var_data, axis=1)
else:
if has_level_size:
time_array[:] = var_data[:, 0, :, :]
else:
time_array[:] = var_data
time_arrays.append(time_array)
dataset = np.stack(time_arrays, axis=-1)
print("processed dataset shape:", dataset.shape)
#processing to time series
default_intervals = [-120, -56, -28, -12, -8, -4, -3, -2, -1, 0, 4]
#todo: handle longer window size
inputs = []
labels = []
for i in range(len(dataset)):
sequence = []
for interval in default_intervals:
index = i+interval
if 0 <= index < len(dataset):
sequence.append(dataset[index])
if len(sequence) == len(default_intervals):
inputs.append(sequence[0:len(sequence)-1])
labels.append([sequence[-1]])
inputs = np.stack(inputs, axis=0)
labels = np.stack(labels, axis=0)
num_samples = len(inputs)
if use_level:
inputs = np.transpose(inputs, (0, 1, 5, 4, 3, 2))
labels = np.transpose(labels, (0, 1, 5, 4, 3, 2))
else:
inputs = np.transpose(inputs, (0, 1, 4, 3, 2))
labels = np.transpose(labels, (0, 1, 4, 3, 2))
flattened_inputs = inputs.reshape(-1, inputs.shape[2])
flattened_labels = labels.reshape(-1, labels.shape[2])
inputs_mean = np.mean(flattened_inputs, axis=0).reshape(1,1,12,1,1)
labels_mean = np.mean(flattened_labels, axis=0).reshape(1,1,12,1,1)
inputs_std = np.std(flattened_inputs, axis=0).reshape(1,1,12,1,1)
labels_std = np.std(flattened_labels, axis=0).reshape(1,1,12,1,1)
normalized_inputs = (inputs - inputs_mean) / inputs_std
normalized_labels = (labels - labels_mean) / labels_std
inputs = normalized_inputs.reshape(inputs.shape)
labels = normalized_labels.reshape(labels.shape)
#keep a copy of original data before splitting and shuffling for sequence prediction testing
original_inputs = torch.tensor(inputs.copy())
original_labels = torch.tensor(labels.copy())
#shuffle
inputs = inputs[torch.randperm(len(inputs), axis=0)]
labels = inputs[torch.randperm(len(labels), axis=0)]
#split into training and testing
training_size = math.floor(split_percentage*num_samples)
X_train = torch.tensor(inputs[0:training_size], dtype=torch.float32)
X_test = torch.tensor(inputs[training_size::], dtype=torch.float32)
Y_train = torch.tensor(labels[0:training_size], dtype=torch.float32)
Y_test = torch.tensor(labels[training_size::], dtype=torch.float32)
# X: (num_samples, sequence_len, features, lat, lon)
# Y: (num_samples, 1, features, lat, lon)
train_dataset = TensorDataset(X_train, Y_train)
test_dataset = TensorDataset(X_test, Y_test)
original_dataset = TensorDataset(original_inputs, original_labels)
torch.save(train_dataset, './data/train_dataset_norm_simple.pth')
torch.save(test_dataset, './data/test_dataset_norm_simple.pth')
torch.save(original_dataset, './data/original_dataset_norm_simple.pth')
return train_dataset, test_dataset
# np.save('../data/test_data_array.npy', dataset)
preprocess_data(0.8)