Skip to main content

​
Class Signature

class ChronosModel(Basemodel):
    def __init__(self, config=None, repo=None)
The ChronosModel class implements Amazon’s Chronos, a language model-based approach to probabilistic time series forecasting that uses tokenization to convert time series into discrete tokens.

​
Initialization Parameters

​
config
dict
default:"None"
Model configuration dictionary. Used when initializing a new model without pre-trained weights.
​
repo
str
default:"None"
Hugging Face model repository ID. If provided, loads the pre-trained Chronos model. If not provided, initializes a new model using the config.

​
Methods

​
finetune()

def finetune(dataset: ChronosDataset, **kwargs)
Finetune the model on the given dataset.
​
dataset
ChronosDataset
required
Dataset for finetuning. Call get_data_loader() to get the dataloader.
​
**kwargs
dict
Optional keyword arguments (currently uses default hyperparameters: lr=1e-4, epochs=5).
​
return
None
The model is finetuned in-place.

​
evaluate()

def evaluate(
    dataset: ChronosDataset,
    horizon_len: int,
    quantile_levels: list,
    metric_only=False,
    **kwargs
)
Evaluate the Chronos model on a dataset.
​
dataset
ChronosDataset
required
Dataset for evaluation. Use get_data_loader() to obtain the dataloader.
​
horizon_len
int
required
Forecast horizon length.
​
quantile_levels
list[float]
required
Quantile levels used to compute quantile forecasts and scoring metrics (e.g., [0.1, 0.5, 0.9]).
​
metric_only
bool
default:"False"
If True, return only a dict of metrics. If False, return metrics plus arrays.
​
return
Dict[str, float] | Tuple
When metric_only=True:Dictionary containing:
  • mse: Mean Squared Error
  • mae: Mean Absolute Error
  • mase: Mean Absolute Scaled Error
  • mape: Mean Absolute Percentage Error
  • rmse: Root Mean Squared Error
  • nrmse: Normalized RMSE
  • smape: Symmetric Mean Absolute Percentage Error
  • msis: Mean Scaled Interval Score
  • nd: Normalized Deviation
  • mwsq: Mean Weighted Scaled Quantile Loss
  • crps: Continuous Ranked Probability Score
When metric_only=False:Tuple of (metrics, trues, preds, histories, quantile_forecasts):
  • metrics: Dictionary of metrics (as above)
  • trues: Ground truth values, shape (num_samples, horizon_len, num_ts)
  • preds: Mean predictions, shape (num_samples, horizon_len, num_ts)
  • histories: Input context sequences, shape (num_samples, context_len, num_ts)
  • quantile_forecasts: Quantile forecasts, shape (num_quantiles, num_samples, horizon_len, num_ts)

​
plot()

def plot(
    dataset: ChronosDataset,
    horizon_len: int,
    quantile_levels: list,
    **kwargs
)
Plot forecast results produced by the Chronos pipeline.
​
dataset
ChronosDataset
required
Dataset for plotting. Use get_data_loader() to obtain the dataloader.
​
horizon_len
int
required
Forecast horizon length.
​
quantile_levels
list[float]
required
List of quantile levels to request from the model (e.g., [0.1, 0.5, 0.9]).
​
**kwargs
dict
Additional keyword arguments forwarded to the visualization helper.
​
return
None
This method does not return a value. It displays visualizations.

​
Usage Example

from samay.model import ChronosModel
from samay.dataset import ChronosDataset

# Load pre-trained model
model = ChronosModel(repo="amazon/chronos-t5-small")

# Or initialize without pre-trained weights
config = {...}  # ChronosConfig parameters
model = ChronosModel(config=config)

# Prepare dataset
dataset = ChronosDataset(...)

# Finetune
model.finetune(dataset)

# Evaluate with quantiles
quantile_levels = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]
metrics = model.evaluate(
    dataset,
    horizon_len=96,
    quantile_levels=quantile_levels,
    metric_only=True
)
print(f"CRPS: {metrics['crps']}")

# Visualize forecasts
model.plot(dataset, horizon_len=96, quantile_levels=quantile_levels)

​
Notes

  • Chronos uses a T5-based architecture to treat forecasting as a sequence-to-sequence problem
  • The model tokenizes time series values into discrete bins before processing
  • Quantile predictions are generated through the model’s probabilistic outputs
  • Three model sizes are available: small, base, and large
  • The model supports variable-length context and prediction horizons

Build docs developers (and LLMs) love

Get started for freeTalk to us