Logging argparsing #34

amangupta2 · 2024-12-18T08:58:29Z

-- Built on branch ablation_10layers
-- Replaced all print statements with logger.info
-- And now parsing command line arguments using argparse. Agreed, it is much neater and convenient.

If all looks good, we can merge this with ablation 10_layers and merge that with main.

…all folders, ann, attention, and TL

… Load one input checkpoints, infer, return output.

…n UNet. Not implemented for TL so far.

… Attention UNets, and Transfer Learning

TomMelt

Thanks @amangupta2 , these changes will greatly improve the scripts and make them easier to run and more robust.

The argparse now adds additional sanity checks to your inputs which will help prevent user error. Great job 👍

I have made a couple of suggestions. We can discuss them in the next meeting if you prefer.

TomMelt · 2025-01-16T11:24:08Z

ann_cnn_training/function_training.py

@@ -119,7 +116,7 @@ def Training_ANN_CNN(


 def Inference_and_Save_ANN_CNN(
-    model, testset, testloader, bs_test, device, stencil, log_filename, outfile
+    model, testset, testloader, bs_test, device, stencil, logger, outfile


FYI, you shouldn't pass the logger as an argument. It should be configured at the file level. This will give better logging information.

Suggested change

model, testset, testloader, bs_test, device, stencil, logger, outfile

model, testset, testloader, bs_test, device, stencil, outfile

Use something like this at the start of each file, e.g., for this specific file it would look like:

from netCDF4 import Dataset import numpy as np import torch import torch.nn as nn import torch.nn.functional as F import torch.optim as optim import logging logger = logging.getLogger(__name__)

TomMelt · 2025-01-16T11:24:26Z

ann_cnn_training/function_training.py

-def Inference_and_Save_AttentionUNet(
-    model, testset, testloader, bs_test, device, log_filename, outfile
-):
+def Inference_and_Save_AttentionUNet(model, testset, testloader, bs_test, device, logger, outfile):


same here

Suggested change

def Inference_and_Save_AttentionUNet(model, testset, testloader, bs_test, device, logger, outfile):

def Inference_and_Save_AttentionUNet(model, testset, testloader, bs_test, device, outfile):

TomMelt · 2025-01-16T11:55:14Z

ann_cnn_training/inference.py

+parser.add_argument("-c", "--ckpt_dir", default=".", help="Checkpoint directory")
+parser.add_argument("-o", "--output_dir", default=".", help="Output directory to save outputs")


for path arguments I would set their type to be path e.g., something like:

Suggested change

parser.add_argument("-c", "--ckpt_dir", default=".", help="Checkpoint directory")

parser.add_argument("-o", "--output_dir", default=".", help="Output directory to save outputs")

from pathlib import Path

parser.add_argument("-c", "--ckpt_dir", default=Path.cwd(), help="Checkpoint directory", type=Path)

parser.add_argument("-o", "--output_dir", default=Path.cwd(), help="Output directory to save outputs", type=Path)

Unfortunately one downside of argparse compared to click is it does not verify paths. So you can still pass in anything.

The benefit to this approach is however, it is platform independent as Path.cwd() should work on Windows too.

Path.cwd() is equivalent to Path("."). It is the path for the Current Working Directory (CWD).

This gives you some extra protection.

TomMelt · 2025-01-16T11:57:23Z

ann_cnn_training/inference.py

+    "-m",
+    "--month",
+    type=int,
+    choices=[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12],


you could simplify this a little:

Suggested change

choices=[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12],

choices=range(1,13),

metavar="{1,2,...,12}",

metavar isn't required but it makes the help output look nicer

TomMelt · 2025-01-16T13:37:01Z

ann_cnn_training/inference_ifs.py

@@ -21,15 +25,65 @@

 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

+parser = argparse.ArgumentParser()


similar changes here to argparse as before for ann_cnn_training/inference.py

TomMelt · 2025-01-16T13:38:47Z

ifs_transfer_learning/attn_inference.py

+    "-m",
+    "--month",
+    type=int,
+    choices=[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12],


similar changes here to argparse as before for ann_cnn_training/inference.py

TomMelt · 2025-01-16T13:39:02Z

ifs_transfer_learning/training_ifs_transfer_learning.py

+    type=int,
+    help="checkpoint (epoch)of the model to be used for transfer learning",
+)
+parser.add_argument(


similar changes here to argparse as before for ann_cnn_training/inference.py

TomMelt · 2025-01-16T13:42:33Z

quick_inference_script/two_sample_inference.py

+# if model == "attention":
+#    test_file = [
+#        f"test_files/test_1x1_inputfeatures_u_v_theta_w_uw_vw_era5_training_data_hourly_2015_constant_mu_sigma_scaling08.nc"
+#    ]
+# elif model == "ann":
+#    if stencil == 1:
+#        test_file = [
+#            f"test_files/test_1x1_inputfeatures_u_v_theta_w_uw_vw_era5_training_data_hourly_2015_constant_mu_sigma_scaling08.nc"
+#        ]
+#    elif stencil == 3:
+#        test_file = [
+#            f"test_files/test_nonlocal_3x3_inputfeatures_u_v_theta_w_uw_vw_era5_training_data_hourly_2015_constant_mu_sigma_scaling08.nc"
+#        ]
+
+


Suggested change

# if model == "attention":

# test_file = [

# f"test_files/test_1x1_inputfeatures_u_v_theta_w_uw_vw_era5_training_data_hourly_2015_constant_mu_sigma_scaling08.nc"

# ]

# elif model == "ann":

# if stencil == 1:

# test_file = [

# f"test_files/test_1x1_inputfeatures_u_v_theta_w_uw_vw_era5_training_data_hourly_2015_constant_mu_sigma_scaling08.nc"

# ]

# elif stencil == 3:

# test_file = [

# f"test_files/test_nonlocal_3x3_inputfeatures_u_v_theta_w_uw_vw_era5_training_data_hourly_2015_constant_mu_sigma_scaling08.nc"

# ]

best practice to remove unused code rather than "commenting" it out. We can always go back to an earlier version (through git/GitHib) if we need it again.

TomMelt · 2025-01-16T13:42:57Z

quick_inference_script/two_sample_inference.py

+    # print(
+    #    f"Model created. \n --- model size: {model.totalsize():.2f} MBs,\n --- Num params: {model.totalparams()/10**6:.3f} mil. "
+    # )


same here. We can either delete it, or move it to the logger

TomMelt · 2025-01-16T13:43:05Z

quick_inference_script/two_sample_inference.py

+    # print(
+    #    f"Model created. \n --- model size: {model.totalsize():.2f} MBs,\n --- Num params: {model.totalparams()/10**6:.3f} mil. "
+    # )


amangupta2 added 4 commits December 13, 2024 14:42

Replaced all custom write_log instances with a predefined logger for …

e306361

…all folders, ann, attention, and TL

updated the quick_inference_script to mimic the CAM coupling process.…

9a9942e

… Load one input checkpoints, infer, return output.

Replaced sys.argv parsing with argparsing for 1. ANNs and 2. Attentio…

b7df62a

…n UNet. Not implemented for TL so far.

Completed adding logging and argparse on all three experiments: ANNs,…

f917a82

… Attention UNets, and Transfer Learning

amangupta2 requested review from TomMelt and j-emberton December 18, 2024 08:58

TomMelt requested changes Jan 16, 2025

View reviewed changes

This was referenced Jan 16, 2025

wip: example refactor with logging and command line options #32

Closed

MWE fortran python #33

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Logging argparsing #34

Logging argparsing #34

amangupta2 commented Dec 18, 2024 •

edited

Loading

TomMelt left a comment

TomMelt Jan 16, 2025 •

edited

Loading

TomMelt Jan 16, 2025

TomMelt Jan 16, 2025

TomMelt Jan 16, 2025

TomMelt Jan 16, 2025

TomMelt Jan 16, 2025

TomMelt Jan 16, 2025

TomMelt Jan 16, 2025

TomMelt Jan 16, 2025

TomMelt Jan 16, 2025

	model, testset, testloader, bs_test, device, stencil, logger, outfile
	model, testset, testloader, bs_test, device, stencil, outfile

	def Inference_and_Save_AttentionUNet(model, testset, testloader, bs_test, device, logger, outfile):
	def Inference_and_Save_AttentionUNet(model, testset, testloader, bs_test, device, outfile):

		parser.add_argument("-c", "--ckpt_dir", default=".", help="Checkpoint directory")
		parser.add_argument("-o", "--output_dir", default=".", help="Output directory to save outputs")

-parser.add_argument("-c", "--ckpt_dir", default=".", help="Checkpoint directory")
-parser.add_argument("-o", "--output_dir", default=".", help="Output directory to save outputs")
+from pathlib import Path
+parser.add_argument("-c", "--ckpt_dir", default=Path.cwd(), help="Checkpoint directory", type=Path)
+parser.add_argument("-o", "--output_dir", default=Path.cwd(), help="Output directory to save outputs", type=Path)

	choices=[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12],
	choices=range(1,13),
	metavar="{1,2,...,12}",

		@@ -21,15 +25,65 @@

		device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

		parser = argparse.ArgumentParser()

Logging argparsing #34

Are you sure you want to change the base?

Logging argparsing #34

Conversation

amangupta2 commented Dec 18, 2024 • edited Loading

TomMelt left a comment

Choose a reason for hiding this comment

TomMelt Jan 16, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

amangupta2 commented Dec 18, 2024 •

edited

Loading

TomMelt Jan 16, 2025 •

edited

Loading