Comments (1)
base config
full_field: &FULL_FIELD
loss: 'l2'
lr: 1E-3
scheduler: 'ReduceLROnPlateau'
num_data_workers: 4
dt: 1 # how many timesteps ahead the model will predict
n_history: 0 #how many previous timesteps to consider
prediction_type: 'iterative'
prediction_length: 41 #applicable only if prediction_type == 'iterative'
n_initial_conditions: 5 #applicable only if prediction_type == 'iterative'
ics_type: "default"
save_raw_forecasts: !!bool True
save_channel: !!bool False
masked_acc: !!bool False
maskpath: None
perturb: !!bool False
add_grid: !!bool False
N_grid_channels: 0
gridtype: 'sinusoidal' #options 'sinusoidal' or 'linear'
roll: !!bool False
max_epochs: 50
batch_size: 64
#afno hyperparams
num_blocks: 8
nettype: 'afno'
patch_size: 8
width: 56
modes: 32
img_size: [720, 1440]
#options default, residual
target: 'default'
in_channels: [0,1]
out_channels: [0,1] #must be same as in_channels if prediction_type == 'iterative'
normalization: 'zscore' #options zscore (minmax not supported)
train_data_path: '/mnt/petrelfs/gongjunchao/h5s' #'/pscratch/sd/j/jpathak/wind/train'
valid_data_path: '/mnt/petrelfs/gongjunchao/h5s' #'/pscratch/sd/j/jpathak/wind/test'
inf_data_path: '/pscratch/sd/j/jpathak/wind/out_of_sample' # test set path for inference
exp_dir: '/mnt/lustre/gongjunchao/FCN/results' #'/pscratch/sd/j/jpathak/ERA5_expts_gtc/wind'
time_means_path: '/pscratch/sd/j/jpathak/wind/time_means.npy'
global_means_path: '/pscratch/sd/j/jpathak/wind/global_means.npy'
global_stds_path: '/pscratch/sd/j/jpathak/wind/global_stds.npy'
orography: !!bool False
orography_path: None
log_to_screen: !!bool True
log_to_wandb: !!bool True
save_checkpoint: !!bool True
enable_nhwc: !!bool False
optimizer_type: 'FusedAdam'
crop_size_x: None
crop_size_y: None
two_step_training: !!bool False
plot_animations: !!bool False
add_noise: !!bool False
noise_std: 0
# type of dataset
ds_type: 'h5'
afno_backbone: &backbone
<<: *FULL_FIELD
log_to_wandb: !!bool False
lr: 5E-4
batch_size: 64 #64
max_epochs: 150
scheduler: 'CosineAnnealingLR'
in_channels: [0, 1 ,2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
out_channels: [0, 1 ,2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
orography: !!bool False
orography_path: None
exp_dir: '/mnt/lustre/gongjunchao/FCN/results' #'/pscratch/sd/j/jpathak/ERA5_expts_gtc/wind'
train_data_path: '/mnt/petrelfs/gongjunchao/h5s' #'/pscratch/sd/j/jpathak/wind/train'
valid_data_path: '/mnt/petrelfs/gongjunchao/h5s' #'/pscratch/sd/j/jpathak/wind/test'
inf_data_path: '/pscratch/sd/s/shas1693/data/era5/out_of_sample'
time_means_path: '/pscratch/sd/s/shas1693/data/era5/time_means.npy'
global_means_path: '/mnt/lustre/gongjunchao/global_means.npy' #'/pscratch/sd/s/shas1693/data/era5/global_means.npy'
global_stds_path: '/mnt/lustre/gongjunchao/global_stds.npy' #'/pscratch/sd/s/shas1693/data/era5/global_stds.npy'
afno_backbone_orography: &backbone_orography
<<: *backbone
orography: !!bool True
orography_path: '/pscratch/sd/s/shas1693/data/era5/static/orography.h5'
afno_backbone_finetune:
<<: *backbone
lr: 1E-4
batch_size: 64
log_to_wandb: !!bool True
max_epochs: 50
pretrained: !!bool True
two_step_training: !!bool True
pretrained_ckpt_path: '/pscratch/sd/s/shas1693/results/era5_wind/afno_backbone/0/training_checkpoints/best_ckpt.tar'
perturbations:
<<: *backbone
lr: 1E-4
batch_size: 64
max_epochs: 50
pretrained: !!bool True
two_step_training: !!bool True
pretrained_ckpt_path: '/pscratch/sd/j/jpathak/ERA5_expts_gtc/wind/afno_20ch_bs_64_lr5em4_blk_8_patch_8_cosine_sched/1/training_checkpoints/best_ckpt.tar'
prediction_length: 24
ics_type: "datetime"
n_perturbations: 100
save_channel: !bool True
save_idx: 4
save_raw_forecasts: !!bool False
date_strings: ["2018-01-01 00:00:00"]
inference_file_tag: " "
valid_data_path: "/pscratch/sd/j/jpathak/ "
perturb: !!bool True
n_level: 0.3
PRECIP
precip: &precip
<<: *backbone
in_channels: [0, 1 ,2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
out_channels: [0]
nettype: 'afno'
nettype_wind: 'afno'
log_to_wandb: !!bool True
lr: 2.5E-4
batch_size: 64
max_epochs: 25
precip: '/pscratch/sd/p/pharring/ERA5/precip/total_precipitation'
time_means_path_tp: '/pscratch/sd/p/pharring/ERA5/precip/total_precipitation/time_means.npy'
model_wind_path: '/pscratch/sd/s/shas1693/results/era5_wind/afno_backbone_finetune/0/training_checkpoints/best_ckpt.tar'
precip_eps: !!float 1e-5
from fourcastnet.
Related Issues (13)
- Checklist before open sourcing
- Possibility of using different resolution input data over smaller areas
- Trouble with downloading data HOT 6
- Help editing code to run train.py on smaller h5 files
- TP accuracy HOT 1
- minor fixes
- the way to calculate `time_means` in script get_stats.py is wrong HOT 3
- AFNO implementation differs from paper
- Pre-processing stage key error HOT 6
- problem with downloading pre-trained model HOT 1
- Pre-processing parallel_copy.py
- How to understand the core code in FFT?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from fourcastnet.