Transfer learning refers to the process of pre-training a flexible model on a large dataset and using it later on other data with little to no training. It is one of the most outstanding 🚀 achievements in Machine Learning and has many practical applications. For time series forecasting, the technique allows you to get lightning-fast predictions ⚡ bypassing the tradeoff between accuracy and speed (more than 30 times faster than our already fast AutoARIMA for a similar accuracy). This notebook shows how to generate a pre-trained model to forecast new time series never seen by the model. Table of Contents
  • Installing MLForecast
  • Load M3 Monthly Data
  • Instantiate NeuralForecast core, Fit, and save
  • Use the pre-trained model to predict on AirPassengers
  • Evaluate Results
You can run these experiments with Google Colab. Open In Colab

​
Installing Libraries

%%capture
# !pip install mlforecast datasetsforecast utilsforecast s3fs

import lightgbm as lgb
import numpy as np
import pandas as pd
from datasetsforecast.m3 import M3
from sklearn.metrics import mean_absolute_error
from utilsforecast.plotting import plot_series

from mlforecast import MLForecast
from mlforecast.target_transforms import Differences

​
Load M3 Data

The M3 class will automatically download the complete M3 dataset and process it. It return three Dataframes: Y_df contains the values for the target variables, X_df contains exogenous calendar features and S_df contains static features for each time-series. For this example we will only use Y_df. If you want to use your own data just replace Y_df. Be sure to use a long format and have a simmilar structure than our data set. 
Y_df_M3, _, _ = M3.load(directory='./', group='Monthly')
In this tutorial we are only using 1_000 series to speed up computations. Remove the filter to use the whole dataset. 
fig = plot_series(Y_df_M3)

fig.savefig('../../figs/transfer_learning__eda.png', bbox_inches='tight')

​
Model Training

Using the MLForecast.fit method you can train a set of models to your dataset. You can modify the hyperparameters of the model to get a better accuracy, in this case we will use the default hyperparameters of lgb.LGBMRegressor. 
models = [lgb.LGBMRegressor(verbosity=-1)]
The MLForecast object has the following parameters:
  • models: a list of sklearn-like (fit and predict) models.
  • freq: a string indicating the frequency of the data. See panda’s available frequencies.
  • differences: Differences to take of the target before computing the features. These are restored at the forecasting step.
  • lags: Lags of the target to use as features.
In this example, we are only using differences and lags to produce features. See the full documentation to see all available features. Any settings are passed into the constructor. Then you call its fit method and pass in the historical data frame Y_df_M3. 
fcst = MLForecast(
    models=models, 
    lags=range(1, 13),
    freq='MS',
    target_transforms=[Differences([1, 12])],
)
fcst.fit(Y_df_M3);

​
Transfer M3 to AirPassengers

Now we can transfer the trained model to forecast AirPassengers with the MLForecast.predict method, we just have to pass the new dataframe to the new_data argument. 
Y_df = pd.read_csv('https://datasets-nixtla.s3.amazonaws.com/air-passengers.csv', parse_dates=['ds'])

# We define the train df. 
Y_train_df = Y_df[Y_df.ds<='1959-12-31'] # 132 train
Y_test_df = Y_df[Y_df.ds>'1959-12-31']   # 12 test

Y_hat_df = fcst.predict(h=12, new_df=Y_train_df)
Y_hat_df.head()
unique_iddsLGBMRegressor
0AirPassengers1960-01-01422.740096
1AirPassengers1960-02-01399.480193
2AirPassengers1960-03-01458.220289
3AirPassengers1960-04-01442.960385
4AirPassengers1960-05-01461.700482
Y_hat_df = Y_test_df.merge(Y_hat_df, how='left', on=['unique_id', 'ds'])

fig = plot_series(Y_train_df, Y_hat_df)

fig.savefig('../../figs/transfer_learning__forecast.png', bbox_inches='tight')

​
Evaluate Results

We evaluate the forecasts of the pre-trained model with the Mean Absolute Error (mae). MAE=1Horizon∑τ∣yτ−y^τ∣\qquad MAE = \frac{1}{Horizon} \sum_{\tau} |y_{\tau} - \hat{y}_{\tau}|\qquad 
y_true = Y_test_df.y.values
y_hat = Y_hat_df['LGBMRegressor'].values

print(f'LGBMRegressor     MAE: {mean_absolute_error(y_hat, y_true):.3f}')
print('ETS               MAE: 16.222')
print('AutoARIMA         MAE: 18.551')

LGBMRegressor     MAE: 13.560
ETS               MAE: 16.222
AutoARIMA         MAE: 18.551