more doc tweaks

README.rst

@@ -1,11 +1,13 @@
 torchcast
 ==========
 
-``torchcast`` is a Python package for forecasting with state-space models built on top of `PyTorch <http://pytorch.org>`_. Its focus is training and forecasting with *batches* of time-series, rather than training separate models for one time-series at a time. In addition, it provides robust support for *multivariate* time-series, where multiple correlated measures are being forecasted.
+``torchcast`` is a Python package for forecasting built on top of `PyTorch <http://pytorch.org>`_. Its benefits include:
 
-Currently the focus of ``torchcast`` is building models that are hybrids: they are classic state-space models with the twist that every part of these models is differentiable and can take advantage of PyTorch's flexibility. For `example <https://torchcast.readthedocs.io/en/latest/examples/electricity.html#Training-our-Hybrid-Forecasting-Model>`_, we can use any  PyTorch ``Module`` to predict the variance of forecasts or even to generate the underlying states themselves.
+1. An API designed around training and forecasting with *batches* of time-series, rather than training separate models for one time-series at a time.
+2. Robust support for *multivariate* time-series, where multiple correlated measures are being forecasted.
+3. Forecasting models that are hybrids: they are classic state-space models with the twist that every part is differentiable and can take advantage of PyTorch's flexibility. For `example <https://docs.strong.io/torchcast/examples/electricity.html#Training-our-Hybrid-Forecasting-Model>`_, we can use arbitrary PyTorch :class:`torch.nn.Modules` to learn seasonal variations across multiple groups, embedding complex seasonality into lower-dimensional space.
 
-This repository is the work of `Strong Analytics <https://www.strong.io/>`_.
+This repository is the work of `Strong <https://www.strong.io/>`_.
 
 .. image:: docs/examples_air_quality_6_2.png
 
@@ -20,4 +22,4 @@ Getting Started
 
 ``torchcast`` requires Python >= 3.8 and PyTorch >= 1.8.
 
-See the `Quick Start <https://torchcast.readthedocs.io/en/latest/quick_start.html>`_ for a simple example that will get you up to speed, or delve into the `examples <https://torchcast.readthedocs.io/en/latest/examples/examples.html>`_ or the `API <https://torchcast.readthedocs.io/en/latest/api/api.html>`_.
+See the `Quick Start <https://docs.strong.io/torchcast/quick_start.html>`_ for a simple example that will get you up to speed, or delve into the `examples <https://docs.strong.io/torchcast/examples/examples.html>`_ or the `API <https://docs.strong.io/torchcast/api/api.html>`_.

docs/api/processes.rst

@@ -5,4 +5,4 @@ Processes
 
 .. automodule:: torchcast.process
    :members: LocalLevel, LocalTrend, Season, LinearModel
-   :exclude-members: forward
+   :exclude-members: forward, solve_and_predict

docs/quick_start.py

@@ -127,7 +127,7 @@
 # The `Predictions` class comes with a `plot` classmethod for getting simple plots of forecasted vs. actual:
 
 # %%
-print(pred.plot(df_pred.query("group=='Changping'"), split_dt=SPLIT_DT, time_colname='time', group_colname='group'))
+pred.plot(df_pred.query("group=='Changping'"), split_dt=SPLIT_DT, time_colname='time', group_colname='group')
 
 # %% [markdown]
 # Finally you can produce dataframes that decompose the predictions into the underlying `processes` that produced them:

pyproject.toml

@@ -25,7 +25,7 @@ dynamic = ["version"]
 [project.urls]
 Repository = "https://github.com/strongio/torchcast"
 "Bug Tracker" = "https://github.com/strongio/torchcast/issues"
-Documentation = "https://torchcast.readthedocs.io/en/latest/"
+Documentation = "https://docs.strong.io/torchcast/"
 
 [project.optional-dependencies]
 dev = [

torchcast/state_space/predictions.py

@@ -11,7 +11,9 @@
 from functools import cached_property
 
 from torchcast.internals.utils import get_nan_groups, is_near_zero, transpose_last_dims
-from torchcast.utils import conf2bounds, TimeSeriesDataset, class_or_instancemethod
+from torchcast.utils import TimeSeriesDataset
+from torchcast.utils.stats import conf2bounds
+from torchcast.utils.misc import class_or_instancemethod
 
 if TYPE_CHECKING:
     from torchcast.state_space import StateSpaceModel

torchcast/utils/__init__.py

@@ -1,14 +1,13 @@
 """
-The data-utils in this module are useful for converting time-series data from a Pandas DataFrame into a PyTorch
-:class:`torch.utils.data.Dataset` and/or :class:`torch.utils.data.DataLoader`. The most common pattern is using the
-``from_dataframe()`` classmethod.
+This module includes:
 
-Additionally, utility functions are provided for handling missing data and adding calendar-features (i.e.
-weekly/daily/yearly season dummy-features that can be passed to any neural-network).
+- Data-utils for converting time-series data from a Pandas DataFrame into a PyTorch :class:`torch.utils.data.Dataset`
+ and/or :class:`torch.utils.data.DataLoader`. The most common pattern is using the ``from_dataframe()`` classmethod.
+- A function for handling implicit missing data
+- A function for adding calendar-features: i.e. weekly/daily/yearly season dummy-features.
 """
 
 from .features import add_season_features
 from .data import TimeSeriesDataset, TimeSeriesDataLoader, complete_times
-from .stats import conf2bounds
-from .misc import class_or_instancemethod
 from .baseline import make_baseline
+from .training import SimpleTrainer, StateSpaceTrainer, SeasonalEmbeddingsTrainer