Forecasting with foundation models¶

Foundation Models (FMs) have recently introduced a major paradigm shift in the field of time series forecasting. Driven by the same architectural breakthroughs that power Large Language Models (LLMs), these models offer powerful generalization capabilities out-of-the-box.

In the context of time series, a Foundation Model is a large-scale neural network (typically based on Transformer architectures) that has been pre-trained on massive and highly diverse datasets spanning multiple domains (e.g., finance, weather, web traffic, and retail).

Models such as AWS Chronos, Google TimesFM2.5, and Salesforce MOIRAI operate on a principle similar to language models: they treat time series values as a sequence of tokens. Because they have already "seen" millions of time series patterns during their pre-training phase, they can identify trends, seasonality, and complex dynamics in new, unseen data without requiring any domain-specific training.

Foundation Models vs. Machine Learning Models¶

Foundation models and traditional machine learning models approach forecasting in fundamentally different ways. Understanding these distinctions is crucial for knowing when and how to deploy each method.

Zero-Shot Prediction

Machine learning models require a training phase. You must fit the model on your historical target data (and exogenous variables) so the algorithm can learn the optimal weights and parameters for your specific time series. Foundation models are capable of zero-shot prediction. Because the model's parameters are already frozen from its massive pre-training phase, it can generate forecasts on your data immediately, without explicitly "learning" from your specific dataset first.

The Role of the fit Method

Machine learning models must be trained: calling .fit() optimizes the model's internal parameters to minimize an error metric on your data. Foundation models, by contrast, arrive pre-trained, their weights are frozen and never updated. Calling .fit() is therefore not a training step; it simply stores the recent historical context (last window of observations, frequency, and scaling factors) so that .predict() has everything it needs at inference time.

Context Window vs. Engineered Lags

Machine learning models rely on explicitly engineered features. In skforecast, you define lags or window_features to create a tabular dataset where past values are used as columns to predict the target. Foundation models rely on a context window. You pass a raw, sequential chunk of recent historical data (e.g., the last 512 observations) directly into the model at inference time. The attention mechanism inside the model automatically decides which past data points are most relevant.

✎ Note

For more details about forecasting with foundation models, visit Forecasting: Principles and Practice, the Pythonic Way.

Foundation Models in skforecast¶

Skforecast's integration is built on two layers. First, FoundationModel acts as a unified wrapper that adapts each model's native API (Chronos-2, TimesFM 2.5, Moirai-2) behind a familiar scikit-learn interface (.fit(), .predict(), .get_params()). Second, ForecasterFoundation wraps that estimator to unlock the full skforecast ecosystem (backtesting, prediction intervals, multi-series support...) with the same code you would use for any other forecaster.

Supported Foundation Models¶

	Chronos	TimesFM	Moirai	TabICL
Provider	Amazon	Google	Salesforce	Soda-Inria
GitHub	chronos-forecasting	timesfm	uni2ts	tabicl
Documentation	Chronos models	TimesFM models	Moirai-R models	TabICL Docs
Available model IDs	amazon/chronos-2 autogluon/chronos-2-small autogluon/chronos-2-synth	google/timesfm-2.5-200m-pytorch	Salesforce/moirai-2.0-R-small	soda-inria/tabicl
Backend	PyTorch	PyTorch	PyTorch	PyTorch
Forecasting type	Zero-shot	Zero-shot	Zero-shot	Zero-shot
Default context_length	8192	512	2048	4096
Max context_length	8192	16384	2048	4096
max_horizon	No hard limit, set via steps at predict time	512	No hard limit, set via steps at predict time	No hard limit, set via steps at predict time
Point forecast	Median (0.5 quantile)	Mean (dedicated output array)	Median (0.5 quantile)	Mean (default, configurable to median)
Covariate support (exog)	Yes	No	No	Yes
cross_learning parameter	Yes (multi-series mode only)	No	No	No
Install command	pip install chronos-forecasting	pip install git+https://github.com/google-research/timesfm.git	pip install uni2ts	pip install tabicl[forecast]

💡 Tip

All three models run on the CPU. However, a CUDA GPU is recommended for faster inference, especially with long context windows. The MPS backend is also detected automatically by PyTorch and can benefit Apple Silicon users.

FoundationModel¶

Each foundation model (Chronos-2, TimesFM 2.5, Moirai-2) has its own API, input format, and inference logic. The FoundationModel class abstracts away these differences behind a single scikit-learn-compatible interface (.fit(), .predict()). Internally, it delegates to a model-specific adapter that handles all the implementation details.

ForecasterFoundation¶

The ForecasterFoundation wraps a FoundationModel and exposes the same interface as any other skforecast forecaster. This means that you can use backtesting, prediction intervals, and multi-series support with the exact same code. It also makes it easy to compare a zero-shot Chronos model with a trained ForecasterRecursive on the same dataset.

Input Data Formats¶

ForecasterFoundation accepts several data formats for both the target series and exogenous variables.

Target Series (series)

The series parameter in the .fit() method supports both single-series and multi-series (global model) configurations.

Mode	Allowed Data Type	Description
Single-Series	pd.Series	A single time series with a named index.
Multi-Series (Wide)	pd.DataFrame	Each column represents a separate time series.
Multi-Series (Long)	pd.DataFrame	MultiIndex (Level 0: series ID, Level 1: DatetimeIndex).
Multi-Series (Dict)	dict[str, pd.Series]	Keys are series identifiers, values are pandas Series.

💡 Tip

While Long-format DataFrames are supported, they are converted to dictionaries internally. For best performance, pass a dict[str, pd.Series] directly.

Exogenous Variables (exog)

Exogenous variables must be aligned with the target series index. Currently, only Chronos supports covariates (see the Supported Foundation Models table). TimesFM 2.5 and Moirai-2 do not accept exogenous variables.

Mode	Allowed Data Type	Description
Single-Series	pd.Series or pd.DataFrame	Aligned to the target series index.
Multi-Series (Dict)	dict[str, pd.Series \| pd.DataFrame \| None]	One entry per series.
Multi-Series (Broadcast)	pd.Series or pd.DataFrame	Automatically applied to all series.
Multi-Series (Long)	pd.DataFrame	MultiIndex (Level 0: series ID, Level 1: DatetimeIndex).

Libraries and data¶

In [1]:

# Libraries
# ==============================================================================
import numpy as np
import pandas as pd
import torch
import matplotlib.pyplot as plt
from skforecast.datasets import fetch_dataset
from skforecast.foundation import FoundationModel, ForecasterFoundation
from skforecast.model_selection import (
    TimeSeriesFold,
    backtesting_foundation
)
from skforecast.plot import set_dark_theme

color = '\033[1m\033[38;5;208m' 
print(f"{color}torch version: {torch.__version__}")
print(f"  Cuda available : {torch.cuda.is_available()}")
print(f"  MPS available  : {torch.backends.mps.is_available()}")

# Libraries # ============================================================================== import numpy as np import pandas as pd import torch import matplotlib.pyplot as plt from skforecast.datasets import fetch_dataset from skforecast.foundation import FoundationModel, ForecasterFoundation from skforecast.model_selection import ( TimeSeriesFold, backtesting_foundation ) from skforecast.plot import set_dark_theme color = '\033[1m\033[38;5;208m' print(f"{color}torch version: {torch.__version__}") print(f" Cuda available : {torch.cuda.is_available()}") print(f" MPS available : {torch.backends.mps.is_available()}")

torch version: 2.4.1
  Cuda available : False
  MPS available  : True

In [2]:

# Data download
# ==============================================================================
data = fetch_dataset(name='vic_electricity')

# Aggregating in 1H intervals
# ==============================================================================
# The Date column is eliminated so that it does not generate an error when aggregating.
data = data.drop(columns="Date")
data = (
    data
    .resample(rule="h", closed="left", label="right")
    .agg({
        "Demand": "mean",
        "Temperature": "mean",
        "Holiday": "mean",
    })
)
data.head(3)

# Data download # ============================================================================== data = fetch_dataset(name='vic_electricity') # Aggregating in 1H intervals # ============================================================================== # The Date column is eliminated so that it does not generate an error when aggregating. data = data.drop(columns="Date") data = ( data .resample(rule="h", closed="left", label="right") .agg({ "Demand": "mean", "Temperature": "mean", "Holiday": "mean", }) ) data.head(3)

╭──────────────────────────── vic_electricity ─────────────────────────────╮
│ Description:                                                             │
│ Half-hourly electricity demand for Victoria, Australia                   │
│                                                                          │
│ Source:                                                                  │
│ O'Hara-Wild M, Hyndman R, Wang E, Godahewa R (2022).tsibbledata: Diverse │
│ Datasets for 'tsibble'. https://tsibbledata.tidyverts.org/,              │
│ https://github.com/tidyverts/tsibbledata/.                               │
│ https://tsibbledata.tidyverts.org/reference/vic_elec.html                │
│                                                                          │
│ URL:                                                                     │
│ https://raw.githubusercontent.com/skforecast/skforecast-                 │
│ datasets/main/data/vic_electricity.csv                                   │
│                                                                          │
│ Shape: 52608 rows x 4 columns                                            │
╰──────────────────────────────────────────────────────────────────────────╯

Out[2]:

	Demand	Temperature	Holiday
Time
2011-12-31 14:00:00	4323.095350	21.225	1.0
2011-12-31 15:00:00	3963.264688	20.625	1.0
2011-12-31 16:00:00	3950.913495	20.325	1.0

In [3]:

# Split data into train-test
# ==============================================================================
data = data.loc['2012-01-01 00:00:00':'2014-12-30 23:00:00', :].copy()
end_train = '2014-11-30 23:59:00'
data_train = data.loc[: end_train, :].copy()
data_test  = data.loc[end_train:, :].copy()

print(f"Train dates: {data_train.index.min()} --- {data_train.index.max()}  (n={len(data_train)})")
print(f"Test dates : {data_test.index.min()} --- {data_test.index.max()}  (n={len(data_test)})")

# Split data into train-test # ============================================================================== data = data.loc['2012-01-01 00:00:00':'2014-12-30 23:00:00', :].copy() end_train = '2014-11-30 23:59:00' data_train = data.loc[: end_train, :].copy() data_test = data.loc[end_train:, :].copy() print(f"Train dates: {data_train.index.min()} --- {data_train.index.max()} (n={len(data_train)})") print(f"Test dates : {data_test.index.min()} --- {data_test.index.max()} (n={len(data_test)})")

Train dates: 2012-01-01 00:00:00 --- 2014-11-30 23:00:00  (n=25560)
Test dates : 2014-12-01 00:00:00 --- 2014-12-30 23:00:00  (n=720)

⚠ Warning: Data Leakage

The following examples use a widely available public dataset for illustrative purposes. It is highly probable that the foundation models (Chronos, TimesFM, Moirai) were exposed to these data points during their pre-training phase. As a result, the predictions may be more optimistic than what would be achieved in a real-world production environment with private or novel data.

Chronos¶

A ForecasterFoundation is created using Amazon's Chronos-2-small model.

In [4]:

# Create ForecasterFoundation
# ==============================================================================
estimator = FoundationModel(model_id="autogluon/chronos-2-small")
forecaster = ForecasterFoundation(estimator=estimator)

# Create ForecasterFoundation # ============================================================================== estimator = FoundationModel(model_id="autogluon/chronos-2-small") forecaster = ForecasterFoundation(estimator=estimator)

Each adapter accepts additional keyword arguments that control model-specific behavior (e.g., context_length, device_map, torch_dtype). These can be passed directly through the FoundationModel constructor.

For the full list of available parameters, see the API reference: ChronosAdapter, TimesFMAdapter, MoiraiAdapter.

💡 Tip

While .fit() is used here to store the historical context and metadata, it is not strictly required. Foundation models can generate forecasts by passing the context directly to .predict() via the context parameter. However, calling .fit() first simplifies subsequent calls to .predict(), .predict_interval(), and .predict_quantiles().

In [5]:

# Train ForecasterFoundation
# ==============================================================================
forecaster.fit(
    series = data_train["Demand"], 
    exog   = data_train[["Temperature", "Holiday"]]
)
forecaster

# Train ForecasterFoundation # ============================================================================== forecaster.fit( series = data_train["Demand"], exog = data_train[["Temperature", "Holiday"]] ) forecaster

Out[5]:

ForecasterFoundation

General Information

• Model ID: autogluon/chronos-2-small
• Context length: 8192
• Window size: 8192
• Series names: Demand
• Exogenous included: True
• Creation date: 2026-04-23 15:33:01
• Last fit date: 2026-04-23 15:33:02
• Skforecast version: 0.22.0
• Python version: 3.12.13
• Forecaster id: None

Exogenous Variables

Temperature, Holiday

Training Information

• Context range: 'Demand': ['2012-01-01 00:00:00', '2014-11-30 23:00:00']
• Training index type: DatetimeIndex
• Training index frequency:

Model Parameters

• cross_learning: False
• context_length: 8192
• device_map: auto
• torch_dtype: None
• predict_kwargs: None

📖 API Reference    📝 User Guide

Three methods can be used to predict the next $n$ steps ahead: predict(), predict_interval(), and predict_quantiles(). All these methods allow for passing context and context_exog to override the historical context used by the underlying model to generate predictions.

In [6]:

# Predictions: point forecast
# ==============================================================================
steps = 24
predictions = forecaster.predict(
                  steps = steps,
                  exog  = data_test[["Temperature", "Holiday"]]
              )

predictions.head(3)

# Predictions: point forecast # ============================================================================== steps = 24 predictions = forecaster.predict( steps = steps, exog = data_test[["Temperature", "Holiday"]] ) predictions.head(3)

Out[6]:

	level	pred
2014-12-01 00:00:00	Demand	5488.940430
2014-12-01 01:00:00	Demand	5431.977539
2014-12-01 02:00:00	Demand	5347.607422

In [7]:

# Predictions: intervals
# ==============================================================================
predictions_intervals = forecaster.predict_interval(
                            steps    = steps,
                            exog     = data_test[["Temperature", "Holiday"]],
                            interval = [10, 90],  # 80% prediction interval
                        )

predictions_intervals.head(3)

# Predictions: intervals # ============================================================================== predictions_intervals = forecaster.predict_interval( steps = steps, exog = data_test[["Temperature", "Holiday"]], interval = [10, 90], # 80% prediction interval ) predictions_intervals.head(3)

Out[7]:

	level	pred	lower_bound	upper_bound
2014-12-01 00:00:00	Demand	5488.940430	5333.779785	5682.048340
2014-12-01 01:00:00	Demand	5431.977539	5262.784180	5657.536133
2014-12-01 02:00:00	Demand	5347.607422	5161.753418	5592.012695

In [8]:

# Backtesting
# ==============================================================================
cv = TimeSeriesFold(
         steps              = 24,
         initial_train_size = len(data.loc[:end_train]),
         refit              = False
     )

metrics_chronos, backtest_predictions = backtesting_foundation(
    forecaster        = forecaster,
    series            = data['Demand'],
    exog              = data[["Temperature", "Holiday"]],
    cv                = cv,
    metric            = 'mean_absolute_error',
    suppress_warnings = True
)

print("Backtest metrics")
display(metrics_chronos)
print("")
print("Backtest predictions")
backtest_predictions.head(4)

# Backtesting # ============================================================================== cv = TimeSeriesFold( steps = 24, initial_train_size = len(data.loc[:end_train]), refit = False ) metrics_chronos, backtest_predictions = backtesting_foundation( forecaster = forecaster, series = data['Demand'], exog = data[["Temperature", "Holiday"]], cv = cv, metric = 'mean_absolute_error', suppress_warnings = True ) print("Backtest metrics") display(metrics_chronos) print("") print("Backtest predictions") backtest_predictions.head(4)

  0%|          | 0/30 [00:00<?, ?it/s]

Backtest metrics

	mean_absolute_error
0	158.693401

Backtest predictions

Out[8]:

	level	fold	pred
2014-12-01 00:00:00	Demand	0	5488.940430
2014-12-01 01:00:00	Demand	0	5431.977539
2014-12-01 02:00:00	Demand	0	5347.607422
2014-12-01 03:00:00	Demand	0	5256.606934

In [9]:

# Plot predictions
# ==============================================================================
set_dark_theme()
fig, ax = plt.subplots(figsize=(7, 3))
data_test['Demand'].plot(ax=ax, label='test')
backtest_predictions['pred'].plot(ax=ax, label='predictions')
ax.legend();

# Plot predictions # ============================================================================== set_dark_theme() fig, ax = plt.subplots(figsize=(7, 3)) data_test['Demand'].plot(ax=ax, label='test') backtest_predictions['pred'].plot(ax=ax, label='predictions') ax.legend();

Multiple series (global model)¶

The class ForecasterFoundation allows modeling and forecasting multiple series with a single model.

In [10]:

# Data
# ==============================================================================
data_multiseries = fetch_dataset(name="items_sales")
display(data_multiseries.head(3))

# Data # ============================================================================== data_multiseries = fetch_dataset(name="items_sales") display(data_multiseries.head(3))

╭─────────────────────── items_sales ───────────────────────╮
│ Description:                                              │
│ Simulated time series for the sales of 3 different items. │
│                                                           │
│ Source:                                                   │
│ Simulated data.                                           │
│                                                           │
│ URL:                                                      │
│ https://raw.githubusercontent.com/skforecast/skforecast-  │
│ datasets/main/data/simulated_items_sales.csv              │
│                                                           │
│ Shape: 1097 rows x 3 columns                              │
╰───────────────────────────────────────────────────────────╯

	item_1	item_2	item_3
date
2012-01-01	8.253175	21.047727	19.429739
2012-01-02	22.777826	26.578125	28.009863
2012-01-03	27.549099	31.751042	32.078922

In [11]:

# Split data into train-test
# ==============================================================================
end_train = '2014-07-15 23:59:00'
data_multiseries_train = data_multiseries.loc[:end_train, :]
data_multiseries_test  = data_multiseries.loc[end_train:, :]

# Split data into train-test # ============================================================================== end_train = '2014-07-15 23:59:00' data_multiseries_train = data_multiseries.loc[:end_train, :] data_multiseries_test = data_multiseries.loc[end_train:, :]

In [12]:

# Plot time series
# ==============================================================================
set_dark_theme()
fig, axes = plt.subplots(nrows=3, ncols=1, figsize=(8, 5), sharex=True)

for i, col in enumerate(data_multiseries.columns):
    data_multiseries_train[col].plot(ax=axes[i], label='train')
    data_multiseries_test[col].plot(ax=axes[i], label='test')
    axes[i].set_title(col)
    axes[i].set_ylabel('sales')
    axes[i].set_xlabel('')
    axes[i].legend(loc='upper left')

fig.tight_layout()
plt.show();

# Plot time series # ============================================================================== set_dark_theme() fig, axes = plt.subplots(nrows=3, ncols=1, figsize=(8, 5), sharex=True) for i, col in enumerate(data_multiseries.columns): data_multiseries_train[col].plot(ax=axes[i], label='train') data_multiseries_test[col].plot(ax=axes[i], label='test') axes[i].set_title(col) axes[i].set_ylabel('sales') axes[i].set_xlabel('') axes[i].legend(loc='upper left') fig.tight_layout() plt.show();

In this example, instead of calling fit(), the context is passed directly to the predict() method.

In [13]:

# Create and train ForecasterFoundation
# ==============================================================================
estimator = FoundationModel(model_id = "autogluon/chronos-2-small")
forecaster = ForecasterFoundation(estimator = estimator)

# fit() is optional; context is passed directly to predict()
# forecaster.fit(series=data_multiseries_train)

forecaster

# Create and train ForecasterFoundation # ============================================================================== estimator = FoundationModel(model_id = "autogluon/chronos-2-small") forecaster = ForecasterFoundation(estimator = estimator) # fit() is optional; context is passed directly to predict() # forecaster.fit(series=data_multiseries_train) forecaster

Out[13]:

ForecasterFoundation

General Information

• Model ID: autogluon/chronos-2-small
• Context length: 8192
• Window size: 8192
• Series names: None
• Exogenous included: False
• Creation date: 2026-04-23 15:33:20
• Last fit date: None
• Skforecast version: 0.22.0
• Python version: 3.12.13
• Forecaster id: None

Exogenous Variables

None

Training Information

• Context range: Not fitted
• Training index type: Not fitted
• Training index frequency: Not fitted

Model Parameters

• cross_learning: False
• context_length: 8192
• device_map: auto
• torch_dtype: None
• predict_kwargs: None

📖 API Reference    📝 User Guide

In [14]:

# Predictions for all series (levels)
# ==============================================================================
steps = 24

predictions_items = forecaster.predict(
                        steps   = len(data_multiseries_test), 
                        levels  = None,  # All levels are predicted
                        context = data_multiseries_train
                    )

predictions_items.head()

# Predictions for all series (levels) # ============================================================================== steps = 24 predictions_items = forecaster.predict( steps = len(data_multiseries_test), levels = None, # All levels are predicted context = data_multiseries_train ) predictions_items.head()

╭────────────────────────────────── InputTypeWarning ──────────────────────────────────╮
│ Passing a DataFrame (either wide or long format) as `series` requires additional     │
│ internal transformations, which can increase computational time. It is recommended   │
│ to use a dictionary of pandas Series instead. For more details, see:                 │
│ https://skforecast.org/latest/user_guides/independent-multi-time-series-forecasting. │
│ html#input-data                                                                      │
│                                                                                      │
│ Category : skforecast.exceptions.InputTypeWarning                                    │
│ Location :                                                                           │
│ /opt/homebrew/Caskroom/miniconda/base/envs/skforecast_py12/lib/python3.12/site-packa │
│ ges/skforecast/utils/utils.py:2799                                                   │
│ Suppress : warnings.simplefilter('ignore', category=InputTypeWarning)                │
╰──────────────────────────────────────────────────────────────────────────────────────╯

Out[14]:

	level	pred
2014-07-16	item_1	25.478813
2014-07-16	item_2	10.479328
2014-07-16	item_3	11.807503
2014-07-17	item_1	25.194397
2014-07-17	item_2	10.714880

In [15]:

# Plot predictions
# ==============================================================================
set_dark_theme()
fig, axes = plt.subplots(nrows=3, ncols=1, figsize=(8, 5), sharex=True)

for i, col in enumerate(data_multiseries.columns):
    
    data_multiseries_train[col].plot(ax=axes[i], label='train')
    data_multiseries_test[col].plot(ax=axes[i], label='test')
    predictions_items.query(f"level == '{col}'").plot(
        ax=axes[i], label='predictions', color='white'
    )

    axes[i].set_title(col)
    axes[i].set_ylabel('sales')
    axes[i].set_xlabel('')
    axes[i].legend(loc='upper left')

fig.tight_layout()
plt.show();

# Plot predictions # ============================================================================== set_dark_theme() fig, axes = plt.subplots(nrows=3, ncols=1, figsize=(8, 5), sharex=True) for i, col in enumerate(data_multiseries.columns): data_multiseries_train[col].plot(ax=axes[i], label='train') data_multiseries_test[col].plot(ax=axes[i], label='test') predictions_items.query(f"level == '{col}'").plot( ax=axes[i], label='predictions', color='white' ) axes[i].set_title(col) axes[i].set_ylabel('sales') axes[i].set_xlabel('') axes[i].legend(loc='upper left') fig.tight_layout() plt.show();

In [16]:

# Interval predictions for item_1 and item_2
# ==============================================================================
predictions_intervals = forecaster.predict_interval(
                            steps    = 24,
                            levels   = ['item_1', 'item_2'],
                            context  = data_multiseries_train,
                            interval = [10, 90],  # 80% prediction interval
                        )

predictions_intervals.head()

# Interval predictions for item_1 and item_2 # ============================================================================== predictions_intervals = forecaster.predict_interval( steps = 24, levels = ['item_1', 'item_2'], context = data_multiseries_train, interval = [10, 90], # 80% prediction interval ) predictions_intervals.head()

╭────────────────────────────────── InputTypeWarning ──────────────────────────────────╮
│ Passing a DataFrame (either wide or long format) as `series` requires additional     │
│ internal transformations, which can increase computational time. It is recommended   │
│ to use a dictionary of pandas Series instead. For more details, see:                 │
│ https://skforecast.org/latest/user_guides/independent-multi-time-series-forecasting. │
│ html#input-data                                                                      │
│                                                                                      │
│ Category : skforecast.exceptions.InputTypeWarning                                    │
│ Location :                                                                           │
│ /opt/homebrew/Caskroom/miniconda/base/envs/skforecast_py12/lib/python3.12/site-packa │
│ ges/skforecast/utils/utils.py:2799                                                   │
│ Suppress : warnings.simplefilter('ignore', category=InputTypeWarning)                │
╰──────────────────────────────────────────────────────────────────────────────────────╯

Out[16]:

	level	pred	lower_bound	upper_bound
2014-07-16	item_1	25.393276	24.595732	26.200489
2014-07-16	item_2	10.694120	8.742786	13.271328
2014-07-17	item_1	25.141916	24.192814	26.036993
2014-07-17	item_2	10.868244	8.804617	13.735439
2014-07-18	item_1	25.119213	24.065229	26.078648

Other foundation models¶

The examples above use the Amazon Chronos model, but the same code structure applies to any other foundation model supported by skforecast. To use a different model, simply change the model_id parameter when instantiating the FoundationModel wrapper. The rest of the forecasting pipeline remains unchanged, allowing you to easily compare different foundation models on your dataset.

TimesFM2.5¶

A ForecasterFoundation is created using Google's TimesFM-2.5-200m model.

In [17]:

# Create ForecasterFoundation
# ==============================================================================
estimator = FoundationModel(model_id="google/timesfm-2.5-200m-pytorch")
forecaster = ForecasterFoundation(estimator = estimator)

# Create ForecasterFoundation # ============================================================================== estimator = FoundationModel(model_id="google/timesfm-2.5-200m-pytorch") forecaster = ForecasterFoundation(estimator = estimator)

In [18]:

# Train ForecasterFoundation
# ==============================================================================
forecaster.fit(series=data_train["Demand"])
forecaster

# Train ForecasterFoundation # ============================================================================== forecaster.fit(series=data_train["Demand"]) forecaster

Out[18]:

ForecasterFoundation

General Information

• Model ID: google/timesfm-2.5-200m-pytorch
• Context length: 512
• Window size: 512
• Series names: Demand
• Exogenous included: False
• Creation date: 2026-04-23 15:33:22
• Last fit date: 2026-04-23 15:33:22
• Skforecast version: 0.22.0
• Python version: 3.12.13
• Forecaster id: None

Exogenous Variables

None

Training Information

• Context range: 'Demand': ['2012-01-01 00:00:00', '2014-11-30 23:00:00']
• Training index type: DatetimeIndex
• Training index frequency:

Model Parameters

• context_length: 512
• max_horizon: 512
• forecast_config_kwargs: None

📖 API Reference    📝 User Guide

In [19]:

# Predictions: point forecast
# ==============================================================================
steps = 24
predictions = forecaster.predict(steps=steps)
predictions.head(3)

# Predictions: point forecast # ============================================================================== steps = 24 predictions = forecaster.predict(steps=steps) predictions.head(3)

Out[19]:

	level	pred
2014-12-01 00:00:00	Demand	5665.909668
2014-12-01 01:00:00	Demand	5668.937012
2014-12-01 02:00:00	Demand	5760.318848

In [20]:

# Predictions: intervals
# ==============================================================================
predictions_intervals = forecaster.predict_interval(
    steps    = steps,
    interval = [10, 90],  # 80% prediction interval
)
predictions_intervals.head(3)

# Predictions: intervals # ============================================================================== predictions_intervals = forecaster.predict_interval( steps = steps, interval = [10, 90], # 80% prediction interval ) predictions_intervals.head(3)

Out[20]:

	level	pred	lower_bound	upper_bound
2014-12-01 00:00:00	Demand	5665.909668	5549.081055	5800.553223
2014-12-01 01:00:00	Demand	5668.937012	5457.991699	5904.702148
2014-12-01 02:00:00	Demand	5760.318848	5443.220703	6093.883789

In [21]:

# Backtesting
# ==============================================================================
cv = TimeSeriesFold(
         steps              = 24,
         initial_train_size = len(data.loc[:end_train]),
         refit              = False
     )

metrics_timesfm, backtest_predictions = backtesting_foundation(
    forecaster        = forecaster,
    series            = data['Demand'],
    cv                = cv,
    metric            = 'mean_absolute_error',
    suppress_warnings = True
)

print("Backtest metrics")
display(metrics_timesfm)
print("")
print("Backtest predictions")
backtest_predictions.head(4)

# Backtesting # ============================================================================== cv = TimeSeriesFold( steps = 24, initial_train_size = len(data.loc[:end_train]), refit = False ) metrics_timesfm, backtest_predictions = backtesting_foundation( forecaster = forecaster, series = data['Demand'], cv = cv, metric = 'mean_absolute_error', suppress_warnings = True ) print("Backtest metrics") display(metrics_timesfm) print("") print("Backtest predictions") backtest_predictions.head(4)

  0%|          | 0/168 [00:00<?, ?it/s]

Backtest metrics

	mean_absolute_error
0	163.088465

Backtest predictions

Out[21]:

	level	fold	pred
2014-07-16 00:00:00	Demand	0	6188.992676
2014-07-16 01:00:00	Demand	0	5991.776855
2014-07-16 02:00:00	Demand	0	5831.969727
2014-07-16 03:00:00	Demand	0	5704.315430

Moirai¶

A ForecasterFoundation is created using Salesforce's Moirai-2.0-R-small model.

In [22]:

# Create ForecasterFoundation
# ==============================================================================
estimator = FoundationModel(model_id="Salesforce/moirai-2.0-R-small")
forecaster = ForecasterFoundation(estimator=estimator)

# Create ForecasterFoundation # ============================================================================== estimator = FoundationModel(model_id="Salesforce/moirai-2.0-R-small") forecaster = ForecasterFoundation(estimator=estimator)

In [23]:

# Train ForecasterFoundation
# ==============================================================================
forecaster.fit(series=data_train["Demand"])
forecaster

# Train ForecasterFoundation # ============================================================================== forecaster.fit(series=data_train["Demand"]) forecaster

Out[23]:

ForecasterFoundation

General Information

• Model ID: Salesforce/moirai-2.0-R-small
• Context length: 2048
• Window size: 2048
• Series names: Demand
• Exogenous included: False
• Creation date: 2026-04-23 15:33:36
• Last fit date: 2026-04-23 15:33:36
• Skforecast version: 0.22.0
• Python version: 3.12.13
• Forecaster id: None

Exogenous Variables

None

Training Information

• Context range: 'Demand': ['2012-01-01 00:00:00', '2014-11-30 23:00:00']
• Training index type: DatetimeIndex
• Training index frequency:

Model Parameters

• context_length: 2048
• device: auto

📖 API Reference    📝 User Guide

In [24]:

# Predictions: point forecast
# ==============================================================================
steps = 24
predictions = forecaster.predict(steps=steps)
predictions.head(3)

# Predictions: point forecast # ============================================================================== steps = 24 predictions = forecaster.predict(steps=steps) predictions.head(3)

/var/folders/wt/8tvn563d5v55nspfbydgqb9r0000gp/T/ipykernel_40162/2957951684.py:4: UserWarning: MPS device is not supported by Moirai because the uni2ts library uses float64 operations internally. Falling back to CPU.
  predictions = forecaster.predict(steps=steps)

Out[24]:

	level	pred
2014-12-01 00:00:00	Demand	5795.251953
2014-12-01 01:00:00	Demand	5986.599121
2014-12-01 02:00:00	Demand	6149.235352

In [25]:

# Predictions: intervals
# ==============================================================================
predictions_intervals = forecaster.predict_interval(
    steps    = steps,
    interval = [10, 90],  # 80% prediction interval
)
predictions_intervals.head(3)

# Predictions: intervals # ============================================================================== predictions_intervals = forecaster.predict_interval( steps = steps, interval = [10, 90], # 80% prediction interval ) predictions_intervals.head(3)

Out[25]:

	level	pred	lower_bound	upper_bound
2014-12-01 00:00:00	Demand	5795.251953	5628.000488	5953.445312
2014-12-01 01:00:00	Demand	5986.599121	5673.974121	6254.291016
2014-12-01 02:00:00	Demand	6149.235352	5728.174805	6538.167969

In [26]:

# Backtesting
# ==============================================================================
cv = TimeSeriesFold(
         steps              = 24,
         initial_train_size = len(data.loc[:end_train]),
         refit              = False
     )

metrics_moirai, backtest_predictions = backtesting_foundation(
    forecaster        = forecaster,
    series            = data['Demand'],
    cv                = cv,
    metric            = 'mean_absolute_error',
    suppress_warnings = True
)

print("Backtest metrics")
display(metrics_moirai)
print("")
print("Backtest predictions")
backtest_predictions.head(4)

# Backtesting # ============================================================================== cv = TimeSeriesFold( steps = 24, initial_train_size = len(data.loc[:end_train]), refit = False ) metrics_moirai, backtest_predictions = backtesting_foundation( forecaster = forecaster, series = data['Demand'], cv = cv, metric = 'mean_absolute_error', suppress_warnings = True ) print("Backtest metrics") display(metrics_moirai) print("") print("Backtest predictions") backtest_predictions.head(4)

  0%|          | 0/168 [00:00<?, ?it/s]

Backtest metrics

	mean_absolute_error
0	154.080703

Backtest predictions

Out[26]:

	level	fold	pred
2014-07-16 00:00:00	Demand	0	6224.538086
2014-07-16 01:00:00	Demand	0	6114.214355
2014-07-16 02:00:00	Demand	0	5991.563965
2014-07-16 03:00:00	Demand	0	5959.375488

TabICL¶

A ForecasterFoundation is created using Soda-Inria's TabICL model.

In this case, the context_length is reduced from the default of 4096 to 72 to speed up inference. For the best performance, use as much context as possible, depending on computational capacity.

In [27]:

# Create ForecasterFoundation
# ==============================================================================
estimator = FoundationModel(model_id="soda-inria/tabicl", context_length=72)
forecaster = ForecasterFoundation(estimator=estimator)

# Create ForecasterFoundation # ============================================================================== estimator = FoundationModel(model_id="soda-inria/tabicl", context_length=72) forecaster = ForecasterFoundation(estimator=estimator)

In [28]:

# Train ForecasterFoundation
# ==============================================================================
forecaster.fit(
    series = data_train["Demand"], 
    exog   = data_train[["Temperature", "Holiday"]]
)
forecaster

# Train ForecasterFoundation # ============================================================================== forecaster.fit( series = data_train["Demand"], exog = data_train[["Temperature", "Holiday"]] ) forecaster

Out[28]:

ForecasterFoundation

General Information

• Model ID: soda-inria/tabicl
• Context length: 72
• Window size: 72
• Series names: Demand
• Exogenous included: True
• Creation date: 2026-04-23 15:33:43
• Last fit date: 2026-04-23 15:33:43
• Skforecast version: 0.22.0
• Python version: 3.12.13
• Forecaster id: None

Exogenous Variables

Temperature, Holiday

Training Information

• Context range: 'Demand': ['2012-01-01 00:00:00', '2014-11-30 23:00:00']
• Training index type: DatetimeIndex
• Training index frequency:

Model Parameters

• context_length: 72
• point_estimate: mean
• tabicl_config: None
• temporal_features: None

📖 API Reference    📝 User Guide

In [29]:

# Predictions: point forecast
# ==============================================================================
steps = 24
predictions = forecaster.predict(
                  steps = steps,
                  exog  = data_test[["Temperature", "Holiday"]]
              )

predictions.head(3)

# Predictions: point forecast # ============================================================================== steps = 24 predictions = forecaster.predict( steps = steps, exog = data_test[["Temperature", "Holiday"]] ) predictions.head(3)

Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.28it/s]

Out[29]:

	level	pred
2014-12-01 00:00:00	Demand	5672.044434
2014-12-01 01:00:00	Demand	5729.508301
2014-12-01 02:00:00	Demand	5774.244629

In [30]:

# Predictions: intervals
# ==============================================================================
predictions_intervals = forecaster.predict_interval(
                            steps    = steps,
                            exog     = data_test[["Temperature", "Holiday"]],
                            interval = [10, 90],  # 80% prediction interval
                        )

predictions_intervals.head(3)

# Predictions: intervals # ============================================================================== predictions_intervals = forecaster.predict_interval( steps = steps, exog = data_test[["Temperature", "Holiday"]], interval = [10, 90], # 80% prediction interval ) predictions_intervals.head(3)

Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.48it/s]

Out[30]:

	level	pred	lower_bound	upper_bound
2014-12-01 00:00:00	Demand	5684.678711	5378.824219	5957.540039
2014-12-01 01:00:00	Demand	5745.275879	5359.950195	6097.788574
2014-12-01 02:00:00	Demand	5808.790039	5236.602051	6276.870117

In [31]:

# Backtesting
# ==============================================================================
cv = TimeSeriesFold(
         steps              = 24,
         initial_train_size = len(data.loc[:end_train]),
         refit              = False
     )

metrics_tabicl, backtest_predictions = backtesting_foundation(
    forecaster        = forecaster,
    series            = data['Demand'],
    exog              = data[["Temperature", "Holiday"]],
    cv                = cv,
    metric            = 'mean_absolute_error',
    suppress_warnings = True
)

print("Backtest metrics")
display(metrics_tabicl)
print("")
print("Backtest predictions")
backtest_predictions.head(4)

# Backtesting # ============================================================================== cv = TimeSeriesFold( steps = 24, initial_train_size = len(data.loc[:end_train]), refit = False ) metrics_tabicl, backtest_predictions = backtesting_foundation( forecaster = forecaster, series = data['Demand'], exog = data[["Temperature", "Holiday"]], cv = cv, metric = 'mean_absolute_error', suppress_warnings = True ) print("Backtest metrics") display(metrics_tabicl) print("") print("Backtest predictions") backtest_predictions.head(4)

  0%|          | 0/168 [00:00<?, ?it/s]

Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.87it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.47it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.24it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.32it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.06it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.32it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.32it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.56it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.57it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.50it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.53it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.33it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.24it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.53it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.56it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.42it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.36it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.38it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.65it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.85it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.46it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.38it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.49it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.47it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.33it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.22it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.31it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.44it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.53it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.37it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.10it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.21it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.38it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.05it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.14it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.03it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.37it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.18it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.29it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.02it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.40it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.36it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.40it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.17it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.88it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.41it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.31it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.53it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.41it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.27it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.27it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.46it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.50it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.29it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.44it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.40it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.22it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.21it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.63it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.47it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.46it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.32it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.38it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.73it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.04it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.77it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.29it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.70it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.77it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.94it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.12it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.22it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.19it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.77it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.94it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.40it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.41it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.21it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.23it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.30it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.31it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.42it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.07it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.16it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.39it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.34it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.34it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.29it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.41it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.29it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.28it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.44it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.32it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.50it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.44it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.41it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.95it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.01it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.13it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.36it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.42it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.21it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.15it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.23it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.17it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.36it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.28it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.24it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.40it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.38it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.21it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.28it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.31it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.35it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.41it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.34it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.41it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.14it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.23it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.46it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.38it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.26it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.47it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.27it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.21it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.40it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.33it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.40it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.41it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.13it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.64it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  1.77it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.37it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.37it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.37it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.15it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.12it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.22it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.52it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.44it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.30it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.42it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.40it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.22it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.29it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.27it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.43it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.46it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.44it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.16it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.36it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.34it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.17it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.31it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.27it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.33it/s]
Predicting time series: 100%|██████████| 1/1 [00:00<00:00,  2.28it/s]

Backtest metrics

	mean_absolute_error
0	291.01068

Backtest predictions

Out[31]:

	level	fold	pred
2014-07-16 00:00:00	Demand	0	6206.785156
2014-07-16 01:00:00	Demand	0	6153.166016
2014-07-16 02:00:00	Demand	0	6103.278809
2014-07-16 03:00:00	Demand	0	6089.614258

Model comparison¶

The following table summarizes the backtesting results (Mean Absolute Error) for the three foundation models on the same dataset.

✎ Note

Chronos-2 is the only model that supports exogenous variables, so its backtesting includes them. TimesFM and Moirai do not use exogenous variables. This difference should be considered when comparing results.

In [32]:

# Comparison of backtesting metrics
# ==============================================================================
comparison = pd.DataFrame({
    "Model": [
        "Chronos-2 (small)*",
        "TimesFM-2.5 (200m)",
        "Moirai-2.0-R (small)",
        "TabICLv2"
    ],
    "mean_absolute_error": [
        metrics_chronos["mean_absolute_error"].iloc[0],
        metrics_timesfm["mean_absolute_error"].iloc[0],
        metrics_moirai["mean_absolute_error"].iloc[0],
        metrics_tabicl["mean_absolute_error"].iloc[0]  # Reduced context length
    ],
})

comparison.style.highlight_min(
    subset="mean_absolute_error", color="green"
).format(precision=4)

# Comparison of backtesting metrics # ============================================================================== comparison = pd.DataFrame({ "Model": [ "Chronos-2 (small)*", "TimesFM-2.5 (200m)", "Moirai-2.0-R (small)", "TabICLv2" ], "mean_absolute_error": [ metrics_chronos["mean_absolute_error"].iloc[0], metrics_timesfm["mean_absolute_error"].iloc[0], metrics_moirai["mean_absolute_error"].iloc[0], metrics_tabicl["mean_absolute_error"].iloc[0] # Reduced context length ], }) comparison.style.highlight_min( subset="mean_absolute_error", color="green" ).format(precision=4)

Out[32]:

	Model	mean_absolute_error
0	Chronos-2 (small)*	158.6934
1	TimesFM-2.5 (200m)	163.0885
2	Moirai-2.0-R (small)	154.0807
3	TabICLv2	291.0107