Comments (1)
Hello again guys.
Today I downloaded the pre-processed data (which is uploaded to the globus) for 2018 for example and sliced it to just 10-time steps. I performed the precipitation inference for this small dataset and the results were very disappointing. The acc was around 0.03 and I dont know where do i made a mistake! I will paste here both results. The first output was performed with preprocessed dataset provided here in GitHub and globus and the second one is the output performed on my custom dataset.
2023-06-27 10:49:01,288 - root - INFO - --------------- Versions ---------------
2023-06-27 10:49:01,300 - root - INFO - git branch: b'* master'
2023-06-27 10:49:01,307 - root - INFO - git hash: b'4cac4678bcf003e2692f3c6528ccb4aa62204f4d'
2023-06-27 10:49:01,307 - root - INFO - Torch: 2.0.1+cu118
2023-06-27 10:49:01,307 - root - INFO - ----------------------------------------
2023-06-27 10:49:01,307 - root - INFO - ------------------ Configuration ------------------
2023-06-27 10:49:01,307 - root - INFO - Configuration file: /content/drive/Shareddrives/Final_FCN/FourCastNet/config/AFNO.yaml
2023-06-27 10:49:01,307 - root - INFO - Configuration name: precip
2023-06-27 10:49:01,308 - root - INFO - in_channels [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
2023-06-27 10:49:01,308 - root - INFO - out_channels [0]
2023-06-27 10:49:01,308 - root - INFO - nettype afno
2023-06-27 10:49:01,308 - root - INFO - nettype_wind afno
2023-06-27 10:49:01,308 - root - INFO - log_to_wandb True
2023-06-27 10:49:01,308 - root - INFO - lr 0.00025
2023-06-27 10:49:01,308 - root - INFO - batch_size 64
2023-06-27 10:49:01,308 - root - INFO - max_epochs 25
2023-06-27 10:49:01,308 - root - INFO - precip /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy
2023-06-27 10:49:01,308 - root - INFO - time_means_path_tp /content/drive/Shareddrives/Final_FCN/stats_v0_globus/time_mean_tp/time_means.npy
2023-06-27 10:49:01,308 - root - INFO - model_wind_path /content/drive/Shareddrives/Final_FCN/backbone.ckpt
2023-06-27 10:49:01,308 - root - INFO - precip_eps 1e-05
2023-06-27 10:49:01,308 - root - INFO - scheduler CosineAnnealingLR
2023-06-27 10:49:01,308 - root - INFO - orography False
2023-06-27 10:49:01,309 - root - INFO - orography_path None
2023-06-27 10:49:01,309 - root - INFO - exp_dir /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/output
2023-06-27 10:49:01,309 - root - INFO - train_data_path /pscratch/sd/s/shas1693/data/era5/train
2023-06-27 10:49:01,309 - root - INFO - valid_data_path /pscratch/sd/s/shas1693/data/era5/test
2023-06-27 10:49:01,309 - root - INFO - inf_data_path /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/backbone
2023-06-27 10:49:01,309 - root - INFO - time_means_path /content/drive/Shareddrives/Final_FCN/stats_v0_globus/time_means.npy
2023-06-27 10:49:01,309 - root - INFO - global_means_path /content/drive/Shareddrives/Final_FCN/stats_v0_globus/global_means.npy
2023-06-27 10:49:01,309 - root - INFO - global_stds_path /content/drive/Shareddrives/Final_FCN/stats_v0_globus/global_stds.npy
2023-06-27 10:49:01,309 - root - INFO - loss l2
2023-06-27 10:49:01,309 - root - INFO - num_data_workers 4
2023-06-27 10:49:01,309 - root - INFO - dt 1
2023-06-27 10:49:01,309 - root - INFO - n_history 0
2023-06-27 10:49:01,309 - root - INFO - prediction_type iterative
2023-06-27 10:49:01,310 - root - INFO - prediction_length 3
2023-06-27 10:49:01,310 - root - INFO - n_initial_conditions 5
2023-06-27 10:49:01,310 - root - INFO - ics_type default
2023-06-27 10:49:01,310 - root - INFO - save_raw_forecasts True
2023-06-27 10:49:01,310 - root - INFO - save_channel False
2023-06-27 10:49:01,310 - root - INFO - masked_acc False
2023-06-27 10:49:01,310 - root - INFO - maskpath None
2023-06-27 10:49:01,310 - root - INFO - perturb False
2023-06-27 10:49:01,310 - root - INFO - add_grid False
2023-06-27 10:49:01,310 - root - INFO - N_grid_channels 0
2023-06-27 10:49:01,310 - root - INFO - gridtype sinusoidal
2023-06-27 10:49:01,310 - root - INFO - roll False
2023-06-27 10:49:01,310 - root - INFO - num_blocks 8
2023-06-27 10:49:01,310 - root - INFO - patch_size 8
2023-06-27 10:49:01,311 - root - INFO - width 56
2023-06-27 10:49:01,311 - root - INFO - modes 32
2023-06-27 10:49:01,311 - root - INFO - target default
2023-06-27 10:49:01,311 - root - INFO - normalization zscore
2023-06-27 10:49:01,311 - root - INFO - log_to_screen True
2023-06-27 10:49:01,311 - root - INFO - save_checkpoint True
2023-06-27 10:49:01,311 - root - INFO - enable_nhwc False
2023-06-27 10:49:01,311 - root - INFO - optimizer_type FusedAdam
2023-06-27 10:49:01,311 - root - INFO - crop_size_x None
2023-06-27 10:49:01,311 - root - INFO - crop_size_y None
2023-06-27 10:49:01,311 - root - INFO - two_step_training False
2023-06-27 10:49:01,311 - root - INFO - plot_animations False
2023-06-27 10:49:01,312 - root - INFO - add_noise False
2023-06-27 10:49:01,312 - root - INFO - noise_std 0
2023-06-27 10:49:01,312 - root - INFO - world_size 1
2023-06-27 10:49:01,312 - root - INFO - global_batch_size 64
2023-06-27 10:49:01,312 - root - INFO - experiment_dir /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/output
2023-06-27 10:49:01,312 - root - INFO - best_checkpoint_path /content/drive/Shareddrives/Final_FCN/precip.ckpt
2023-06-27 10:49:01,312 - root - INFO - resuming False
2023-06-27 10:49:01,312 - root - INFO - local_rank 0
2023-06-27 10:49:01,312 - root - INFO - ---------------------------------------------------
2023-06-27 10:49:01,312 - root - INFO - Inference for 1 initial conditions
2023-06-27 10:49:01,315 - root - INFO - Getting file stats from /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/backbone/2011_11.h5
2023-06-27 10:49:01,315 - root - INFO - Number of samples per year: 120
2023-06-27 10:49:01,315 - root - INFO - Found data at path /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/backbone. Number of examples: 120. Image Shape: 720 x 1440 x 20
2023-06-27 10:49:01,316 - root - INFO - Delta t: 6 hours
2023-06-27 10:49:01,316 - root - INFO - Including 0 hours of past history in training at a frequency of 6 hours
2023-06-27 10:49:01,316 - root - INFO - Loading trained model checkpoint from /content/drive/Shareddrives/Final_FCN/precip.ckpt
2023-06-27 10:49:05,853 - root - INFO - Loading validation data
2023-06-27 10:49:05,853 - root - INFO - Validation data from /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/backbone/2011_11.h5
2023-06-27 10:49:05,855 - root - INFO - Loading validation precip data
2023-06-27 10:49:05,855 - root - INFO - Validation data from /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/2018.h5
2023-06-27 10:49:06,239 - root - INFO - Initial condition 1 of 1
2023-06-27 10:49:07,653 - root - INFO - Begin autoregressive inference
2023-06-27 10:49:09,671 - root - INFO - Timestep 0 of 3. TP RMS Error: 0.0, ACC: 1.0
2023-06-27 10:49:12,681 - root - INFO - Timestep 1 of 3. TP RMS Error: 0.0030234134756028652, ACC: 0.031174656003713608
2023-06-27 10:49:14,035 - root - INFO - Timestep 2 of 3. TP RMS Error: 0.002889520488679409, ACC: 0.03148724511265755
2023-06-27 10:49:14,041 - root - INFO - Saving files at /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/output/autoregressive_predictions_tp.h5
2023-06-27 10:49:52,721 - root - INFO - --------------- Versions ---------------
2023-06-27 10:49:52,733 - root - INFO - git branch: b'* master'
2023-06-27 10:49:52,741 - root - INFO - git hash: b'4cac4678bcf003e2692f3c6528ccb4aa62204f4d'
2023-06-27 10:49:52,741 - root - INFO - Torch: 2.0.1+cu118
2023-06-27 10:49:52,741 - root - INFO - ----------------------------------------
2023-06-27 10:49:52,742 - root - INFO - ------------------ Configuration ------------------
2023-06-27 10:49:52,742 - root - INFO - Configuration file: /content/drive/Shareddrives/Final_FCN/FourCastNet/config/AFNO.yaml
2023-06-27 10:49:52,742 - root - INFO - Configuration name: precip
2023-06-27 10:49:52,742 - root - INFO - in_channels [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19]
2023-06-27 10:49:52,742 - root - INFO - out_channels [0]
2023-06-27 10:49:52,742 - root - INFO - nettype afno
2023-06-27 10:49:52,742 - root - INFO - nettype_wind afno
2023-06-27 10:49:52,742 - root - INFO - log_to_wandb True
2023-06-27 10:49:52,742 - root - INFO - lr 0.00025
2023-06-27 10:49:52,742 - root - INFO - batch_size 64
2023-06-27 10:49:52,743 - root - INFO - max_epochs 25
2023-06-27 10:49:52,743 - root - INFO - precip /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy
2023-06-27 10:49:52,743 - root - INFO - time_means_path_tp /content/drive/Shareddrives/Final_FCN/stats_v0_globus/time_mean_tp/time_means.npy
2023-06-27 10:49:52,743 - root - INFO - model_wind_path /content/drive/Shareddrives/Final_FCN/backbone.ckpt
2023-06-27 10:49:52,743 - root - INFO - precip_eps 1e-05
2023-06-27 10:49:52,743 - root - INFO - scheduler CosineAnnealingLR
2023-06-27 10:49:52,743 - root - INFO - orography False
2023-06-27 10:49:52,743 - root - INFO - orography_path None
2023-06-27 10:49:52,743 - root - INFO - exp_dir /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/output
2023-06-27 10:49:52,743 - root - INFO - train_data_path /pscratch/sd/s/shas1693/data/era5/train
2023-06-27 10:49:52,743 - root - INFO - valid_data_path /pscratch/sd/s/shas1693/data/era5/test
2023-06-27 10:49:52,743 - root - INFO - inf_data_path /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/backbone
2023-06-27 10:49:52,744 - root - INFO - time_means_path /content/drive/Shareddrives/Final_FCN/stats_v0_globus/time_means.npy
2023-06-27 10:49:52,744 - root - INFO - global_means_path /content/drive/Shareddrives/Final_FCN/stats_v0_globus/global_means.npy
2023-06-27 10:49:52,744 - root - INFO - global_stds_path /content/drive/Shareddrives/Final_FCN/stats_v0_globus/global_stds.npy
2023-06-27 10:49:52,744 - root - INFO - loss l2
2023-06-27 10:49:52,744 - root - INFO - num_data_workers 4
2023-06-27 10:49:52,744 - root - INFO - dt 1
2023-06-27 10:49:52,744 - root - INFO - n_history 0
2023-06-27 10:49:52,744 - root - INFO - prediction_type iterative
2023-06-27 10:49:52,744 - root - INFO - prediction_length 3
2023-06-27 10:49:52,744 - root - INFO - n_initial_conditions 5
2023-06-27 10:49:52,745 - root - INFO - ics_type default
2023-06-27 10:49:52,745 - root - INFO - save_raw_forecasts True
2023-06-27 10:49:52,745 - root - INFO - save_channel False
2023-06-27 10:49:52,745 - root - INFO - masked_acc False
2023-06-27 10:49:52,745 - root - INFO - maskpath None
2023-06-27 10:49:52,745 - root - INFO - perturb False
2023-06-27 10:49:52,745 - root - INFO - add_grid False
2023-06-27 10:49:52,745 - root - INFO - N_grid_channels 0
2023-06-27 10:49:52,745 - root - INFO - gridtype sinusoidal
2023-06-27 10:49:52,745 - root - INFO - roll False
2023-06-27 10:49:52,745 - root - INFO - num_blocks 8
2023-06-27 10:49:52,746 - root - INFO - patch_size 8
2023-06-27 10:49:52,746 - root - INFO - width 56
2023-06-27 10:49:52,746 - root - INFO - modes 32
2023-06-27 10:49:52,746 - root - INFO - target default
2023-06-27 10:49:52,746 - root - INFO - normalization zscore
2023-06-27 10:49:52,746 - root - INFO - log_to_screen True
2023-06-27 10:49:52,746 - root - INFO - save_checkpoint True
2023-06-27 10:49:52,746 - root - INFO - enable_nhwc False
2023-06-27 10:49:52,746 - root - INFO - optimizer_type FusedAdam
2023-06-27 10:49:52,746 - root - INFO - crop_size_x None
2023-06-27 10:49:52,746 - root - INFO - crop_size_y None
2023-06-27 10:49:52,746 - root - INFO - two_step_training False
2023-06-27 10:49:52,747 - root - INFO - plot_animations False
2023-06-27 10:49:52,747 - root - INFO - add_noise False
2023-06-27 10:49:52,747 - root - INFO - noise_std 0
2023-06-27 10:49:52,747 - root - INFO - world_size 1
2023-06-27 10:49:52,747 - root - INFO - global_batch_size 64
2023-06-27 10:49:52,747 - root - INFO - experiment_dir /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/output
2023-06-27 10:49:52,747 - root - INFO - best_checkpoint_path /content/drive/Shareddrives/Final_FCN/precip.ckpt
2023-06-27 10:49:52,747 - root - INFO - resuming False
2023-06-27 10:49:52,747 - root - INFO - local_rank 0
2023-06-27 10:49:52,747 - root - INFO - ---------------------------------------------------
2023-06-27 10:49:52,747 - root - INFO - Inference for 1 initial conditions
2023-06-27 10:49:52,766 - root - INFO - Getting file stats from /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/backbone/2011_11.h5
2023-06-27 10:49:52,767 - root - INFO - Number of samples per year: 120
2023-06-27 10:49:52,767 - root - INFO - Found data at path /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/backbone. Number of examples: 120. Image Shape: 720 x 1440 x 20
2023-06-27 10:49:52,767 - root - INFO - Delta t: 6 hours
2023-06-27 10:49:52,767 - root - INFO - Including 0 hours of past history in training at a frequency of 6 hours
2023-06-27 10:49:52,768 - root - INFO - Loading trained model checkpoint from /content/drive/Shareddrives/Final_FCN/precip.ckpt
2023-06-27 10:49:57,272 - root - INFO - Loading validation data
2023-06-27 10:49:57,272 - root - INFO - Validation data from /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/backbone/2011_11.h5
2023-06-27 10:49:57,275 - root - INFO - Loading validation precip data
2023-06-27 10:49:57,275 - root - INFO - Validation data from /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/2011_11_tp.h5
2023-06-27 10:49:57,694 - root - INFO - Initial condition 1 of 1
2023-06-27 10:49:58,841 - root - INFO - Begin autoregressive inference
2023-06-27 10:50:00,844 - root - INFO - Timestep 0 of 3. TP RMS Error: 0.0, ACC: 1.0
2023-06-27 10:50:03,975 - root - INFO - Timestep 1 of 3. TP RMS Error: 0.0016368868527933955, ACC: 0.35275188088417053
2023-06-27 10:50:05,338 - root - INFO - Timestep 2 of 3. TP RMS Error: 0.001560143311508, ACC: 0.3554510176181793
2023-06-27 10:50:05,345 - root - INFO - Saving files at /content/drive/Shareddrives/Final_FCN/validation_tp_increase_accuracy/output/autoregressive_predictions_tp.h5
from fourcastnet.
Related Issues (15)
- 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
- Unable to download the weights HOT 1
- Smaller version of dataset
- minor fixes
- the way to calculate `time_means` in script get_stats.py is wrong HOT 3
- how to reproduce your result on 80*40 resolution features? HOT 1
- 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.