params

The params module handles configuration parameters for Nenya models and processing.

Classes

class nenya.params.Params(json_path)[source]

A class for loading and managing hyperparameters from a JSON file.

Parameters:: json_path (str) – Path to the JSON configuration file

save(json_path)[source]

Save parameters to a JSON file.

Parameters:: json_path (str) – Path where the JSON file will be saved

update(json_path)[source]

Update parameters from a JSON file.

Parameters:: json_path (str) – Path to the JSON configuration file

property dict

Returns the parameters as a dictionary.

Returns:: Dictionary of parameters
Return type:: dict

Functions

nenya.params.option_preprocess(opt)[source]

Process options and set derived values.

This function sets up output folders (e.g., model_folder, latents_folder) and processes other options. The object is modified in place.

Parameters:: opt (Params) – Options object to preprocess

Example Usage

from nenya import params

# Load parameters from a JSON file
opt = params.Params("opts_nenya_modis_v5.json")

# Preprocess options
params.option_preprocess(opt)

# Access parameters
learning_rate = opt.learning_rate
batch_size = opt.batch_size_train

# Save parameters to a new file
opt.save("opts_new.json")

Parameter Structure

Example parameter file structure:

{
  "ssl_method": "SimCLR",
  "ssl_model": "resnet50",
  "learning_rate": 0.05,
  "weight_decay": 0.0001,
  "batch_size_train": 64,
  "batch_size_valid": 64,
  "temp": 0.07,
  "trial": 0,
  "cosine": true,
  "warm": true,
  "epochs": 200,
  "model_root": "v5",
  "feat_dim": 128,
  "random_jitter": [5, 5],
  "images_file": "MODIS_2012_96clear_64x64.h5",
  "s3_outdir": "s3://bucket/path",
  "data_folder": "/path/to/data",
  "train_key": "train",
  "valid_key": "valid",
  "cuda_use": true,
  "valid_freq": 5,
  "save_freq": 10
}