Feature/ci

README.md

@@ -1,2 +1,27 @@
 # littlebooktimeseries
+
 Timeseries in 3 hours
+
+## Environment setup
+
+Choose one of the options below to get the workshop notebooks ready.
+
+### Poetry
+- Install Poetry: `curl -sSL https://install.python-poetry.org | python3 -`
+- Install dependencies: `poetry install`
+- Start the shell: `poetry shell`
+- Launch Jupyter: `jupyter lab`
+
+### pip
+- Create a virtual environment: `python3 -m venv .venv`
+- Activate it: `source .venv/bin/activate`
+- Install dependencies: `pip install -r requirements.txt`
+- Launch Jupyter: `jupyter lab`
+
+### uv
+- Install uv: `pip install uv`
+- Sync dependencies: `uv sync`
+- Activate the environment: `source .venv/bin/activate`
+- Launch Jupyter: `jupyter lab`
+
+> All options require Python 3.11 or 3.12. The notebooks live under `notebooks/`; open them in Jupyter Lab during the session.

book/content/pt/part2/hierarchical_forecasting.qmd

@@ -105,7 +105,7 @@ from tsbook.forecasting.reduction import ReductionForecaster
 from lightgbm import LGBMRegressor
 
 forecaster = ReductionForecaster(
-    LGBMRegressor(n_estimators=50, verbose=-1, objective="tweedie"),
+    LGBMRegressor(n_estimators=100, verbose=-1, objective="tweedie", random_state=42),
     window_length=30,
     normalization_strategy="divide_mean",
 )
@@ -296,7 +296,7 @@ pd.DataFrame(
         "OptimalReconciler (ols)": metric(y_test, y_pred_optimal, y_train=y_train),
         "Mint Reconciler": metric(y_test, y_pred_mint, y_train=y_train),
     },
-    index=["Mean Absolute Scaled Error"],
+    index=["Mean Absolute Squared Error"],
 )
 ```
 

book/content/pt/part2/ml_models.qmd

@@ -109,10 +109,10 @@ Primeiro, vamos import `ReductionForecaster`, que é a classe que implementa a a
 
 ```{python}
 from tsbook.forecasting.reduction import ReductionForecaster
-from sklearn.ensemble import RandomForestRegressor
+from lightgbm import LGBMRegressor
 
 model = ReductionForecaster(
-    RandomForestRegressor(n_estimators=100, random_state=42),
+    LGBMRegressor(n_estimators=100, random_state=42),
     window_length=30,
 )
 
@@ -135,7 +135,7 @@ Uma solução é usar a diferenciação para remover a tendência da série.
 ```{python}
 from sktime.transformations.series.difference import Differencer
 
-regressor = RandomForestRegressor(n_estimators=100, random_state=42)
+regressor = LGBMRegressor(n_estimators=100, random_state=42)
 
 model = Differencer() * ReductionForecaster(
     regressor,

book/content/pt/part2/panel_data.qmd

@@ -165,7 +165,7 @@ from tsbook.forecasting.reduction import ReductionForecaster
 from lightgbm import LGBMRegressor
 
 global_forecaster1 = ReductionForecaster(
-    LGBMRegressor(n_estimators=100, verbose=-1),
+    LGBMRegressor(n_estimators=100, verbose=-1, random_state=42),
     window_length=30,
 )
 
@@ -219,7 +219,6 @@ Sabemos como preprocessar séries temporais univariadas para melhorar o desempen
 ```{python}
 from sktime.transformations.series.difference import Differencer
 
-
 global_forecaster2 = Differencer() * global_forecaster1
 global_forecaster2.fit(y_train, X_train)
 ```
@@ -249,7 +248,7 @@ errors["Local (2)"] = metric_local2
 errors
 ```
 
-Note como já superamos o modelo global incial, e o modelo local. Isso é para
+Note como já superamos o modelo global incial. Isso é para
 destacar que é **essencial** realizar um bom preprocessamento e engenharia de features para que modelos de Machine Learning tenham bom desempenho em dados em painel.
 
 ### Normalização por janela