Merge pull request #266 from unit8co/develop

Develop

.github/workflows/develop.yml

@@ -71,7 +71,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        example-name: [FFT-examples.ipynb, RNN-examples.ipynb, darts-intro.ipynb, multivariate-examples.ipynb, data-processing.ipynb]
+        example-name: [04-FFT-examples.ipynb, 05-RNN-examples.ipynb, 01-darts-intro.ipynb, 03-data-processing.ipynb, 02-multi-time-series-and-covariates.ipynb]
     steps:
       - name: "1. Clone repository"
         uses: actions/checkout@v2

.github/workflows/merge.yml

@@ -43,7 +43,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        example-name: [FFT-examples.ipynb, TCN-examples.ipynb, RNN-examples.ipynb, darts-intro.ipynb, multivariate-examples.ipynb, Transformer-examples.ipynb, data-processing.ipynb]
+        example-name: [01-darts-intro.ipynb, 02-multi-time-series-and-covariates.ipynb, 03-data-processing.ipynb, 04-FFT-examples.ipynb, 05-RNN-examples.ipynb]
     steps:
       - name: "1. Clone repository"
         uses: actions/checkout@v2

CHANGELOG.md

@@ -4,7 +4,43 @@ Darts is still in an early development phase and we cannot always guarantee back
 
 ## [Unreleased](https://github.com/unit8co/darts/tree/develop)
 
-[Full Changelog](https://github.com/unit8co/darts/compare/0.5.0...develop)
+[Full Changelog](https://github.com/unit8co/darts/compare/0.6.0...develop)
+
+
+## [0.6.0](https://github.com/unit8co/darts/tree/0.6.0) (2021-02-02)
+
+[Full Changelog](https://github.com/unit8co/darts/compare/0.5.0...0.6.0)
+### For users of the library:
+**Added:**
+- `Pipeline.invertible()` a getter which returns whether the pipeline is invertible or not.
+- `TimeSeries.to_json()` and `TimeSeries.from_json()` methods to convert `TimeSeries` to/from a `JSON` string.
+- New base class `GlobalForecastingModel` for all models supporting training on multiple time series, as well
+as covariates. All PyTorch models are now `GlobalForecastingModel`s.
+- As a consequence of the above, the `fit()` function of PyTorch models (all neural networks) can optionally be called
+with a sequence of time series (instead of a single time series).
+- Similarly, the `predict()` function of these models also accepts a specification of which series should be forecasted
+- A new `TrainingDataset` base class.
+- Some implementations of `TrainingDataset` containing some slicing logic for the training of neural networks on
+several time series.
+- A new `TimeSeriesInferenceDataset` base class.
+- An implementation `SimpleInferenceDataset` of `TimeSeriesInferenceDataset`.
+- All PyTorch models have a new `fit_from_dataset()` method which allows to directly fit the model from a specified
+`TrainingDataset` instance (instead of using a default instance when going via the `fit()` method).
+- A new explanatory notebooks for global models: 
+https://github.com/unit8co/darts/blob/master/examples/02-multi-time-series-and-covariates.ipynb
+
+**Changed:**
+🔴 removed the arguments `training_series` and `target_series` in `ForecastingModel`s. Please consult
+the API documentation of forecasting models to see the new signatures.
+🔴 removed `UnivariateForecastingModel` and `MultivariateForecastingModel` base classes. This distinction does
+not exist anymore. Instead, now some models are "global" (can be trained on multiple series) or "local" (they cannot).
+All implementations of `GlobalForecastingModel`s support multivariate time series out of the box, except N-BEATS.
+- Improved the documentation and README.
+- Re-ordered the example notebooks to improve the flow of examples.
+
+**Fixed:**
+- Many small bug fixes.
+- Unit test speedup by about 15x.
 
 ## [0.5.0](https://github.com/unit8co/darts/tree/0.5.0) (2020-11-09)
 

README.md

@@ -15,20 +15,25 @@ It contains a variety of models, from classics such as ARIMA to neural networks.
 The models can all be used in the same way, using `fit()` and `predict()` functions,
 similar to scikit-learn. The library also makes it easy to backtest models,
 and combine the predictions of several models and external regressors. Darts supports both
-univariate and multivariate time series and models.
+univariate and multivariate time series and models, and the neural networks can be trained
+multiple time series.
+
+## Documentation
+* [Examples & Tutorials](https://unit8co.github.io/darts/examples.html)
+* [API Documentation](https://unit8co.github.io/darts/generated_api/darts.html)
+
+##### High Level Introductions
+* [Introductory Blog Post](https://medium.com/unit8-machine-learning-publication/darts-time-series-made-easy-in-python-5ac2947a8878)
+* [Introductory Video](https://www.youtube.com/watch?v=Sx-uI-PypmU&t=8s&ab_channel=Unit8)
 
 ## Install
 
 We recommend to first setup a clean python environment for your project with at least python 3.6 using your favorite tool ([conda](https://docs.conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html "conda-env"), [venv](https://docs.python.org/3/library/venv.html), [virtualenv](https://virtualenv.pypa.io/en/latest/) with or without [virtualenvwrapper](https://virtualenvwrapper.readthedocs.io/en/latest/)).
 
-### Quick Install
-
-Once your environment is setup you can install darts using the pip package:
+Once your environment is setup you can install darts using pip:
 
     pip install 'u8darts[all]'
 
-### Step-by-step Install
-
 For more detailed install instructions you can refer to our installation guide at the end of this page.
 
 ## Example Usage
@@ -70,11 +75,9 @@ plt.xlabel('Year')
 <img src="https://github.com/unit8co/darts/raw/develop/static/images/example.png" alt="darts forecast example" />
 </div>
 
-We invite you to go over the example notebooks in the `examples` directory.
-
-## Documentation
+We invite you to go over the example and tutorial notebooks in 
+the [examples](https://github.com/unit8co/darts/tree/master/examples) directory.
 
-The documentation of the API and models is available [here](https://unit8co.github.io/darts/).
 
 ## Features
 

darts/__init__.py

@@ -5,28 +5,8 @@
 
 from .timeseries import TimeSeries
 import matplotlib as mpl
-import matplotlib.pyplot as plt
 from matplotlib import cycler
 
-# Enums
-from enum import Enum
-
-
-class SeasonalityMode(Enum):
-    MULTIPLICATIVE = 'multiplicative'
-    ADDITIVE = 'additive'
-    NONE = None
-
-
-class TrendMode(Enum):
-    LINEAR = 'linear'
-    EXPONENTIAL = 'exponential'
-
-
-class ModelMode(Enum):
-    MULTIPLICATIVE = 'multiplicative'
-    ADDITIVE = 'additive'
-
 
 __version__ = '0.5.0'
 

darts/dataprocessing/pipeline.py

@@ -125,6 +125,18 @@ def inverse_transform(self, data: TimeSeries) -> TimeSeries:
             data = transformer.inverse_transform(data)
         return data
 
+    def invertible(self) -> bool:
+        """
+        Returns whether the pipeline is invertible or not.
+        A pipeline is invertible if all transformers in the pipeline are themselves invertible.
+
+        Returns
+        -------
+        bool
+            True if the pipeline is invertible, False otherwise
+        """
+        return self._invertible
+
     def __getitem__(self, key: Union[int, slice]) -> 'Pipeline':
         """
         Gets subset of Pipeline based either on index or slice with indexes.

darts/metrics/metrics.py

@@ -69,6 +69,17 @@ def _get_values_or_raise(series_a: TimeSeries,
     return series_a_common.univariate_values(), series_b_common.univariate_values()
 
 
+def _remove_nan_union(array_a: np.ndarray,
+                      array_b: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
+    """
+    Returnes the two inputs arrays where all elements are deleted that have an index that corresponds to
+    a NaN value in either of the two input arrays.
+    """
+
+    isnan_mask = np.logical_or(np.isnan(array_a), np.isnan(array_b))
+    return np.delete(array_a, isnan_mask), np.delete(array_b, isnan_mask)
+
+
 @multivariate_support
 def mae(series1: TimeSeries,
         series2: TimeSeries,
@@ -100,6 +111,7 @@ def mae(series1: TimeSeries,
     """
 
     y1, y2 = _get_values_or_raise(series1, series2, intersect)
+    y1, y2 = _remove_nan_union(y1, y2)
     return np.mean(np.abs(y1 - y2))
 
 
@@ -134,6 +146,7 @@ def mse(series1: TimeSeries,
     """
 
     y_true, y_pred = _get_values_or_raise(series1, series2, intersect)
+    y_true, y_pred = _remove_nan_union(y_true, y_pred)
     return np.mean((y_true - y_pred)**2)
 
 
@@ -202,6 +215,7 @@ def rmsle(series1: TimeSeries,
     """
 
     y1, y2 = _get_values_or_raise(series1, series2, intersect)
+    y1, y2 = _remove_nan_union(y1, y2)
     y1, y2 = np.log(y1 + 1), np.log(y2 + 1)
     return np.sqrt(np.mean((y1 - y2)**2))
 
@@ -283,6 +297,7 @@ def mape(actual_series: TimeSeries,
     """
 
     y_true, y_hat = _get_values_or_raise(actual_series, pred_series, intersect)
+    y_true, y_hat = _remove_nan_union(y_true, y_hat)
     raise_if_not((y_true != 0).all(), 'The actual series must be strictly positive to compute the MAPE.', logger)
     return 100. * np.mean(np.abs((y_true - y_hat) / y_true))
 
@@ -329,9 +344,10 @@ def smape(actual_series: TimeSeries,
     """
 
     y_true, y_hat = _get_values_or_raise(actual_series, pred_series, intersect)
+    y_true, y_hat = _remove_nan_union(y_true, y_hat)
     raise_if_not(np.logical_or(y_true != 0, y_hat != 0).all(),
                  'The actual series must be strictly positive to compute the sMAPE.', logger)
-    return 200. * np.mean(np.abs((y_true - y_hat) / (np.abs(y_true) + np.abs(y_hat))))
+    return 200. * np.mean(np.abs(y_true - y_hat) / (np.abs(y_true) + np.abs(y_hat)))
 
 
 @multivariate_support
@@ -431,6 +447,7 @@ def ope(actual_series: TimeSeries,
     """
 
     y_true, y_pred = _get_values_or_raise(actual_series, pred_series, intersect)
+    y_true, y_pred = _remove_nan_union(y_true, y_pred)
     y_true_sum, y_pred_sum = np.sum(y_true), np.sum(y_pred)
     raise_if_not(y_true_sum > 0, 'The series of actual value cannot sum to zero when computing OPE.', logger)
     return np.abs((y_true_sum - y_pred_sum) / y_true_sum) * 100.
@@ -474,6 +491,7 @@ def marre(actual_series: TimeSeries,
     """
 
     y_true, y_hat = _get_values_or_raise(actual_series, pred_series, intersect)
+    y_true, y_hat = _remove_nan_union(y_true, y_hat)
     raise_if_not(y_true.max() > y_true.min(), 'The difference between the max and min values must be strictly'
                  'positive to compute the MARRE.', logger)
     true_range = y_true.max() - y_true.min()
@@ -512,6 +530,7 @@ def r2_score(series1: TimeSeries,
     """
 
     y1, y2 = _get_values_or_raise(series1, series2, intersect)
+    y1, y2 = _remove_nan_union(y1, y2)
     ss_errors = np.sum((y1 - y2) ** 2)
     y_hat = y1.mean()
     ss_tot = np.sum((y1 - y_hat) ** 2)

darts/models/arima.py

@@ -13,14 +13,14 @@
 from statsmodels.tsa.arima_model import ARMA as staARMA
 from statsmodels.tsa.arima_model import ARIMA as staARIMA
 
-from .forecasting_model import UnivariateForecastingModel
+from .forecasting_model import ForecastingModel
 from ..timeseries import TimeSeries
 from ..logging import get_logger
 
 logger = get_logger(__name__)
 
 
-class ARIMA(UnivariateForecastingModel):
+class ARIMA(ForecastingModel):
     def __init__(self, p: int = 12, d: int = 1, q: int = 0):
         """ ARIMA
 

darts/models/auto_arima.py

@@ -5,14 +5,14 @@
 
 from pmdarima import AutoARIMA as PmdAutoARIMA
 
-from .forecasting_model import UnivariateForecastingModel
+from .forecasting_model import ForecastingModel
 from ..timeseries import TimeSeries
 from ..logging import get_logger
 
 logger = get_logger(__name__)
 
 
-class AutoARIMA(UnivariateForecastingModel):
+class AutoARIMA(ForecastingModel):
     def __init__(self, *autoarima_args, **autoarima_kwargs):
         """ Auto-ARIMA
 

darts/models/baselines.py

@@ -2,21 +2,21 @@
 Baseline Models
 ---------------
 
-A collection of simple benchmark models.
+A collection of simple benchmark models for univariate series.
 """
 
 from typing import List, Optional
 import numpy as np
 
-from .forecasting_model import ForecastingModel, UnivariateForecastingModel
+from .forecasting_model import ForecastingModel
 from .ensemble_model import EnsembleModel
 from ..timeseries import TimeSeries
 from ..logging import raise_if_not, get_logger
 
 logger = get_logger(__name__)
 
 
-class NaiveMean(UnivariateForecastingModel):
+class NaiveMean(ForecastingModel):
     def __init__(self):
         """ Naive Mean Model
 
@@ -39,7 +39,7 @@ def predict(self, n: int):
         return self._build_forecast_series(forecast)
 
 
-class NaiveSeasonal(UnivariateForecastingModel):
+class NaiveSeasonal(ForecastingModel):
     def __init__(self, K: int = 1):
         """ Naive Seasonal Model
 
@@ -74,7 +74,7 @@ def predict(self, n: int):
         return self._build_forecast_series(forecast)
 
 
-class NaiveDrift(UnivariateForecastingModel):
+class NaiveDrift(ForecastingModel):
     def __init__(self):
         """ Naive Drift Model
 
@@ -112,13 +112,10 @@ def __init__(self, models: List[ForecastingModel]):
         super().__init__(models)
 
     def fit(self, training_series: TimeSeries, target_series: Optional[TimeSeries] = None) -> None:
-        super().fit(training_series, target_series)
+        super().fit(training_series)
 
         for model in self.models:
-            if isinstance(model, UnivariateForecastingModel):
-                model.fit(self.training_series)
-            else:
-                model.fit(self.training_series, self.target_series)
+            model.fit(self.training_series)
 
     def ensemble(self, predictions: List[TimeSeries]):
         return sum(predictions) / len(self.models)

darts/models/ensemble_model.py

@@ -1,6 +1,6 @@
 """
-Ensemble model
---------------
+Ensemble Model Base Class
+-------------------------
 """
 
 from abc import abstractmethod
@@ -34,13 +34,13 @@ def __init__(self, models: List[ForecastingModel]):
         super().__init__()
         self.models = models
 
-    def fit(self, training_series: TimeSeries, target_series: Optional[TimeSeries] = None) -> None:
+    def fit(self, training_series: TimeSeries) -> None:
         """
         Fits the model on the provided series.
         Note that `EnsembleModel.fit()` does NOT call `fit()` on each of its constituent forecasting models.
         It is left to classes inheriting from EnsembleModel to do so appropriately when overriding `fit()`
         """
-        super().fit(training_series, target_series)
+        super().fit(training_series)
 
     def predict(self, n: int) -> TimeSeries:
         super().predict(n)

darts/models/exponential_smoothing.py

@@ -6,15 +6,15 @@
 from typing import Optional
 import statsmodels.tsa.holtwinters as hw
 
-from .forecasting_model import UnivariateForecastingModel
+from .forecasting_model import ForecastingModel
 from ..logging import get_logger
 from ..timeseries import TimeSeries
-from .. import ModelMode
+from ..utils.utils import ModelMode
 
 logger = get_logger(__name__)
 
 
-class ExponentialSmoothing(UnivariateForecastingModel):
+class ExponentialSmoothing(ForecastingModel):
     def __init__(self,
                  trend: Optional[ModelMode] = ModelMode.ADDITIVE,
                  damped: Optional[bool] = False,