File size: 1,643 Bytes
950fc23 | 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 | project:
name: PrecipDD
format_version: precipdd_v1
seed: 23
license: Apache-2.0
paths:
data: data/daily_precip_anomaly.npz
checkpoint: result/checkpoints/precipdd.pt
training_metrics: result/training/metrics.json
predictions: result/output/predictions.npz
evaluation_metrics: result/evaluation/metrics.json
comparison_figure: result/evaluation/comparison.png
data:
layout: NCHW
input_shape: [1, 55, 160]
latitude_range: [-60.0, 75.0]
latitude_note: 55 engineering grid centers at 2.5 degree spacing; the paper text reports inconsistent endpoints alongside 55 points
longitude_range: [0.0, 397.5]
longitude_note: 0-360 E plus repeated 0-40 E extension
normalization: daily climatology anomaly divided by zonal-mean anomaly standard deviation at each latitude
train_samples: 96
validation_samples: 32
test_days_per_year: 6
test_years: [1980, 2020]
synthetic_members: 8
paper_members: 80
paper_period: [1850, 2100]
paper_train_days: 730000
paper_validation_days: 146000
model:
filters: [8, 8, 16, 16, 16]
kernel_size: 3
dense_units: 32
conv_activation: tanh
dense_activation: sigmoid
output_activation: linear
expected_feature_shape: [16, 14, 40]
flattened_features: 8960
training:
ensemble_members: 2
paper_ensemble_members: 5
epochs: 3
batch_size: 8
learning_rate: 0.001
l2_weight_decay: 0.0001
loss: MAE
optimizer: Adam
initialization: Xavier-uniform
num_workers: 0
runtime:
device: auto
ddp_backend_gpu: nccl
ddp_backend_cpu: gloo
evaluation:
emergence_threshold_c: 0.42
occlusion_patch: 7
occlusion_stride: 8
occlusion_max_days: 12
|