Time-series forecasting (Exponential Smoothing) with Darts Integration

i18n/en.yml

@@ -486,6 +486,7 @@ en:
   stocks/pred/lstm: Long-Short Term Memory
   stocks/pred/conv1d: 1D Convolutional Neural Network
   stocks/pred/mc: Monte-Carlo simulations
+  stocks/pred/expo: Probabilistic Exponential Smoothing
   crypto/load: load a specific cryptocurrency for analysis
   crypto/find: find coins
   crypto/_coin: Coin

openbb_terminal/README.md

@@ -186,17 +186,41 @@ Several features in this project utilize Machine Learning. Machine Learning Pyth
 
    To enable the `prediction` menu install additional dependencies after installing main dependencies:
 
-   - On M1 mac
+   - On Intel mac
 
      ```bash
      conda install -c conda-forge tensorflow==2.7.0
+     conda install -c conda-forge -c pytorch u8darts-torch
      poetry install -E prediction
      ```
 
+   - On M1 mac (make sure you are using miniconda3 or miniforge3)
+
+     ```bash
+     brew install cmake
+     brew install gcc
+     conda install -c apple tensorflow-deps
+     conda install -c apple tensorflow-deps==2.8.0
+     python -m pip install tensorflow-metal
+     conda update pip
+     conda update -c defaults numpy
+     export GRPC_PYTHON_BUILD_SYSTEM_OPENSSL=1
+     export GRPC_PYTHON_BUILD_SYSTEM_ZLIB=1
+     wget https://raw.githubusercontent.com/Homebrew/homebrew-core/fb8323f2b170bd4ae97e1bac9bf3e2983af3fdb0/Formula/libomp.rb
+     brew unlink libomp
+     brew install libomp.rb
+     pip3 install -r requirements.txt
+     conda install -c conda-forge tensorflow==2.8.0
+     conda install -c conda-forge -c pytorch u8darts-torch
+     python3 -c "import tensorflow as tf; print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
+     ```
+
    - On all other systems
 
      ```bash
      poetry install -E prediction
+     conda install -c conda-forge -c pytorch u8darts-torch
+
      ```
 
    If you are having trouble with Poetry (e.g. on a Windows system), simply install requirements.txt with pip

openbb_terminal/common/prediction_techniques/ets_view.py

@@ -43,7 +43,7 @@ def display_exponential_smoothing(
     n_predict: int,
     trend: str = "N",
     seasonal: str = "N",
-    seasonal_periods: int = 5,
+    seasonal_periods: int = 7,
     s_end_date: str = "",
     export: str = "",
     time_res: str = "",

openbb_terminal/common/prediction_techniques/expo_model.py

@@ -0,0 +1,137 @@
+"""Probabilistic Exponential Smoothing Model"""
+__docformat__ = "numpy"
+
+import logging
+from typing import Any, Tuple, Union
+
+import numpy as np
+import pandas as pd
+from darts import TimeSeries
+from darts.models import ExponentialSmoothing
+from darts.dataprocessing.transformers import MissingValuesFiller
+from darts.utils.utils import ModelMode, SeasonalityMode
+from darts.metrics import mape
+
+from openbb_terminal.decorators import log_start_end
+from openbb_terminal.rich_config import console
+
+
+TRENDS = ["N", "A", "M"]
+SEASONS = ["N", "A", "M"]
+PERIODS = [4, 5, 7]
+DAMPEN = ["T", "F"]
+
+logger = logging.getLogger(__name__)
+
+
+@log_start_end(log=logger)
+def get_expo_data(
+    data: Union[pd.Series, pd.DataFrame],
+    trend: str = "A",
+    seasonal: str = "A",
+    seasonal_periods: int = 7,
+    dampen: str = "F",
+    n_predict: int = 30,
+    start_window: float = 0.65,
+    forecast_horizon: int = 3,
+) -> Tuple[Any, Any, Any, Any, Any]:
+
+    """Performs Probabilistic Exponential Smoothing forecasting
+    This is a wrapper around statsmodels Holt-Winters' Exponential Smoothing;
+    we refer to this link for the original and more complete documentation of the parameters.
+
+    https://unit8co.github.io/darts/generated_api/darts.models.forecasting.exponential_smoothing.html?highlight=exponential
+
+    Parameters
+    ----------
+    data : Union[pd.Series, np.ndarray]
+        Input data.
+    trend: str
+        Trend component.  One of [N, A, M]
+        Defaults to ADDITIVE.
+    seasonal: str
+        Seasonal component.  One of [N, A, M]
+        Defaults to ADDITIVE.
+    seasonal_periods: int
+        Number of seasonal periods in a year (7 for daily data)
+        If not set, inferred from frequency of the series.
+    dampen: str
+        Dampen the function
+    n_predict: int
+        Number of days to forecast
+    start_window: float
+        Size of sliding window from start of timeseries and onwards
+    forecast_horizon: int
+        Number of days to forecast when backtesting and retraining historical
+
+    Returns
+    -------
+    List[float]
+        Adjusted Data series
+    List[float]
+        List of predicted values
+    Any
+        Fit Prob. Expo model object.
+    """
+
+    filler = MissingValuesFiller()
+    data["date"] = data.index  # add temp column since we need to use index col for date
+    ticker_series = TimeSeries.from_dataframe(
+        data,
+        time_col="date",
+        value_cols=["AdjClose"],
+        freq="B",
+        fill_missing_dates=True,
+    )
+
+    ticker_series = filler.transform(ticker_series)
+    ticker_series = ticker_series.astype(np.float32)
+    _, val = ticker_series.split_before(0.85)
+
+    if trend == "M":
+        trend = ModelMode.MULTIPLICATIVE
+    elif trend == "N":
+        trend = ModelMode.NONE
+    else:  # Default
+        trend = ModelMode.ADDITIVE
+
+    if seasonal == "M":
+        seasonal = SeasonalityMode.MULTIPLICATIVE
+    elif seasonal == "N":
+        seasonal = SeasonalityMode.NONE
+    else:  # Default
+        seasonal = SeasonalityMode.ADDITIVE
+
+    damped = True
+    if dampen == "F":
+        damped = False
+
+    # Model Init
+    model_es = ExponentialSmoothing(
+        trend=trend,
+        seasonal=seasonal,
+        seasonal_periods=int(seasonal_periods),
+        damped=damped,
+        random_state=42,
+    )
+
+    # Training model based on historical backtesting
+    historical_fcast_es = model_es.historical_forecasts(
+        ticker_series,
+        start=float(start_window),
+        forecast_horizon=int(forecast_horizon),
+        verbose=True,
+    )
+
+    # Show forecast over validation # and then +n_predict afterwards sampled 10 times per point
+    probabilistic_forecast = model_es.predict(int(n_predict), num_samples=500)
+    precision = mape(val, probabilistic_forecast)  # mape = mean average precision error
+    console.print(f"model {model_es} obtains MAPE: {precision:.2f}% \n")  # TODO
+
+    return (
+        ticker_series,
+        historical_fcast_es,
+        probabilistic_forecast,
+        precision,
+        model_es,
+    )

openbb_terminal/common/prediction_techniques/expo_view.py

@@ -0,0 +1,118 @@
+"""Probabilistic Exponential Smoothing View"""
+__docformat__ = "numpy"
+
+import logging
+import os
+from typing import Union
+
+import matplotlib.pyplot as plt
+import pandas as pd
+
+from openbb_terminal.config_terminal import theme
+from openbb_terminal.common.prediction_techniques import expo_model
+from openbb_terminal.config_plot import PLOT_DPI
+from openbb_terminal.decorators import log_start_end
+from openbb_terminal.helper_funcs import (
+    export_data,
+    plot_autoscale,
+)
+from openbb_terminal.rich_config import console
+from openbb_terminal.common.prediction_techniques.pred_helper import (
+    print_pretty_prediction,
+)
+
+logger = logging.getLogger(__name__)
+# pylint: disable=too-many-arguments
+
+
+@log_start_end(log=logger)
+def display_expo_forecast(
+    data: Union[pd.DataFrame, pd.Series],
+    ticker_name: str,
+    trend: str,
+    seasonal: str,
+    seasonal_periods: int,
+    dampen: str,
+    n_predict: int,
+    start_window: float,
+    forecast_horizon: int,
+    export: str = "",
+):
+    """Display Probabilistic Exponential Smoothing forecast
+
+    Parameters
+    ----------
+    data : Union[pd.Series, np.array]
+        Data to forecast
+    trend: str
+        Trend component.  One of [N, A, M]
+        Defaults to ADDITIVE.
+    seasonal: str
+        Seasonal component.  One of [N, A, M]
+        Defaults to ADDITIVE.
+    seasonal_periods: int
+        Number of seasonal periods in a year
+        If not set, inferred from frequency of the series.
+    dampen: str
+        Dampen the function
+    n_predict: int
+        Number of days to forecast
+    start_window: float
+        Size of sliding window from start of timeseries and onwards
+    forecast_horizon: int
+        Number of days to forecast when backtesting and retraining historical
+    export: str
+        Format to export data
+    external_axes : Optional[List[plt.Axes]], optional
+        External axes (2 axis is expected in the list), by default None
+    """
+    (
+        ticker_series,
+        historical_fcast_es,
+        predicted_values,
+        precision,
+        _,
+    ) = expo_model.get_expo_data(
+        data,
+        trend,
+        seasonal,
+        seasonal_periods,
+        dampen,
+        n_predict,
+        start_window,
+        forecast_horizon,
+    )
+
+    # Plotting with Matplotlib
+    external_axes = None
+    if not external_axes:
+        fig, ax = plt.subplots(figsize=plot_autoscale(), dpi=PLOT_DPI)
+    else:
+        if len(external_axes) != 1:
+            logger.error("Expected list of one axis item.")
+            console.print("[red]Expected list of one axis item.\n[/red]")
+            return
+        ax = external_axes
+
+    # ax = fig.get_axes()[0] # fig gives list of axes (only one for this case)
+    ticker_series.plot(label="Actual AdjClose", figure=fig)
+    historical_fcast_es.plot(
+        label="Back-test 3-Days ahead forecast (Exp. Smoothing)", figure=fig
+    )
+    predicted_values.plot(
+        label="Probabilistic Forecast", low_quantile=0.1, high_quantile=0.9, figure=fig
+    )
+    ax.set_title(
+        f"PES for ${ticker_name} for next [{n_predict}] days (Model MAPE={round(precision,2)}%)"
+    )
+    ax.set_ylabel("Adj. Closing")
+    ax.set_xlabel("Date")
+    theme.style_primary_axis(ax)
+
+    if not external_axes:
+        theme.visualize_output()
+
+    numeric_forecast = predicted_values.quantile_df()["AdjClose_0.5"].tail(n_predict)
+    print_pretty_prediction(numeric_forecast, data["AdjClose"].iloc[-1])
+
+    export_data(export, os.path.dirname(os.path.abspath(__file__)), "expo")

openbb_terminal/common/prediction_techniques/pred_helper.py

@@ -557,7 +557,6 @@ def lambda_price_prediction_color(val: float, last_val: float) -> str:
 
 def print_pretty_prediction(df_pred: pd.DataFrame, last_price: float):
     """Print predictions"""
-    console.print("")
     if rich_config.USE_COLOR:
         df_pred = pd.DataFrame(df_pred)
         df_pred.columns = ["pred"]

openbb_terminal/stocks/prediction_techniques/pred_api.py

@@ -35,5 +35,9 @@
     display_mc_forecast as mc,
 )
 
+from openbb_terminal.common.prediction_techniques.expo_view import (
+    display_expo_forecast as expo,
+)
+
 # Models
 models = _models(os.path.abspath(os.path.dirname(prediction_techniques.__file__)))