Config loader

Functions to load and store models and datasets.

Code using omegaconf to handle IO.

IO

---

source

class_to_str

def class_to_str(
    cls
):

---

source

load_config

def load_config(
    file_path
):

---

source

config_to_dict

def config_to_dict(
    config
):

---

source

save_dataclass_yaml

def save_dataclass_yaml(
    data_obj, file_path
):

---

source

save_dict_yaml

def save_dict_yaml(
    dict_obj, file_path
):

Test

@dataclass
class MyConfig:    
    target:str = class_to_str(OmegaConf)
    clr_dim: int = 80
    features: list[int]=None
    
c = MyConfig()
c.features = [1,2,3]

OmegaConf.structured(c)

{'target': 'omegaconf.omegaconf.OmegaConf', 'clr_dim': 80, 'features': [1, 2, 3]}

Object config load

Adapted from: https://github.com/Stability-AI/generative-models

---

source

get_obj_from_str

def get_obj_from_str(
    string, reload:bool=False, invalidate_cache:bool=True
):

---

source

instantiate_from_config

def instantiate_from_config(
    config
):

Models

---

source

store_model_state_dict

def store_model_state_dict(
    state_dict, save_path
):

---

source

load_model_state_dict

def load_model_state_dict(
    save_path, device
):

Tensors and numpy

torch.serialization.DEFAULT_PROTOCOL

2

---

source

store_tensor

def store_tensor(
    tensor, save_path, type:str='tensor'
):

---

source

load_tensor

def load_tensor(
    save_path, device, type:str='tensor'
):