Skip to main content

​
Class Signature

class Basemodel:
    def __init__(self, config=None, repo=None)
The Basemodel class serves as the abstract base class for all time series foundation models in Samay. All model implementations inherit from this class and must implement its abstract methods.

​
Initialization Parameters

​
config
dict
default:"None"
Model configuration dictionary containing model-specific parameters.
​
repo
str
default:"None"
Hugging Face model repository ID for loading pre-trained models.

​
Attributes

​
config
dict
Stores the model configuration passed during initialization.
​
device
torch.device
The device (CPU or GPU) on which the model will run. Automatically selects the least used GPU if available, otherwise defaults to CPU.

​
Methods

​
finetune()

def finetune(dataset: BaseDataset, **kwargs)
Finetune the model on the given dataset. This is an abstract method that concrete model classes must implement.
​
dataset
BaseDataset
required
Dataset used for finetuning.
​
**kwargs
dict
Optional keyword arguments for finetuning:
  • lr (float): Learning rate
  • epoch (int): Number of training epochs
  • Additional model-specific parameters
​
return
Any
Model or training result produced by the finetune operation. The exact return type is defined by concrete model implementations.
Raises:
  • NotImplementedError: If a subclass does not implement this method.

​
forecast()

def forecast(input, **kwargs)
Generate forecast(s) from input data.
​
input
Any
required
Input data for forecasting (e.g., torch.Tensor, numpy array, or model-specific input structure).
​
**kwargs
dict
Optional keyword arguments for forecasting (model-specific).
​
return
Any
Forecast outputs. The concrete model defines the exact format (for example, mean forecasts, quantiles, or full distribution).
Raises:
  • NotImplementedError: If a subclass does not implement this method.

​
evaluate()

def evaluate(dataset: BaseDataset, **kwargs)
Evaluate the model on a dataset.
​
dataset
BaseDataset
required
Dataset used for evaluation.
​
**kwargs
dict
Optional evaluation arguments:
  • metric_only (bool): If True, return only metrics dictionary
  • Additional model-specific parameters
​
return
Any
Evaluation metrics or (metrics, details) tuple as defined by the concrete model implementation.
Raises:
  • NotImplementedError: If a subclass does not implement this method.

​
save()

def save(path)
Save the model to disk.
​
path
str
required
Filesystem path where model artifacts should be saved.
​
return
None
This method does not return a value.
Raises:
  • NotImplementedError: If a subclass does not implement this method.

​
Usage Example

from samay.model import Basemodel

# Basemodel is abstract and cannot be instantiated directly
# Use specific model implementations like ChronosModel, MomentModel, etc.

​
Notes

  • The Basemodel class automatically detects and uses the least utilized GPU when available
  • All concrete model implementations must override the abstract methods: finetune(), forecast(), evaluate(), and save()
  • Device selection happens automatically during initialization based on GPU availability

Build docs developers (and LLMs) love

Get started for freeTalk to us