> ## Documentation Index > Fetch the complete documentation index at: https://nixtla-old-docs.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. > Fit an LSTM and NHITS model # Quickstart This notebook provides an example on how to start using the main functionalities of the NeuralForecast library. The [`NeuralForecast`](https://nixtlaverse.nixtla.io/neuralforecast/core.html#neuralforecast) class allows users to easily interact with `NeuralForecast.models` PyTorch models. In this example we will forecast AirPassengers data with a classic [`LSTM`](https://nixtlaverse.nixtla.io/neuralforecast/models.lstm.html#lstm) and the recent [`NHITS`](https://nixtlaverse.nixtla.io/neuralforecast/models.nhits.html#nhits) models. The full list of available models is available [here](https://nixtlaverse.nixtla.io/neuralforecast/docs/capabilities/overview.html). You can run these experiments using GPU with Google Colab. Open In Colab ## 1. Installing NeuralForecast ```python theme={null} !pip install neuralforecast ``` ## 2. Loading AirPassengers Data The `core.NeuralForecast` class contains shared, `fit`, `predict` and other methods that take as inputs pandas DataFrames with columns `['unique_id', 'ds', 'y']`, where `unique_id` identifies individual time series from the dataset, `ds` is the date, and `y` is the target variable. In this example dataset consists of a set of a single series, but you can easily fit your model to larger datasets in long format. ```python theme={null} from neuralforecast.utils import AirPassengersDF ``` ```python theme={null} Y_df = AirPassengersDF Y_df.head() ``` | | unique\_id | ds | y | | - | ---------- | ---------- | ----- | | 0 | 1.0 | 1949-01-31 | 112.0 | | 1 | 1.0 | 1949-02-28 | 118.0 | | 2 | 1.0 | 1949-03-31 | 132.0 | | 3 | 1.0 | 1949-04-30 | 129.0 | | 4 | 1.0 | 1949-05-31 | 121.0 | > **Important** > > DataFrames must include all `['unique_id', 'ds', 'y']` columns. Make > sure `y` column does not have missing or non-numeric values. ## 3. Model Training ### Fit the models Using the [`NeuralForecast.fit`](https://nixtlaverse.nixtla.io/neuralforecast/core.html#neuralforecast.fit) method you can train a set of models to your dataset. You can define the forecasting `horizon` (12 in this example), and modify the hyperparameters of the model. For example, for the [`LSTM`](https://nixtlaverse.nixtla.io/neuralforecast/models.lstm.html#lstm) we changed the default hidden size for both encoder and decoders. ```python theme={null} import logging from neuralforecast import NeuralForecast from neuralforecast.models import LSTM, NHITS, RNN ``` ```python theme={null} logging.getLogger('pytorch_lightning').setLevel(logging.ERROR) ``` ```python theme={null} horizon = 12 # Try different hyperparmeters to improve accuracy. models = [LSTM(input_size=2 * horizon, h=horizon, # Forecast horizon max_steps=500, # Number of steps to train scaler_type='standard', # Type of scaler to normalize data encoder_hidden_size=64, # Defines the size of the hidden state of the LSTM decoder_hidden_size=64,), # Defines the number of hidden units of each layer of the MLP decoder NHITS(h=horizon, # Forecast horizon input_size=2 * horizon, # Length of input sequence max_steps=100, # Number of steps to train n_freq_downsample=[2, 1, 1]) # Downsampling factors for each stack output ] nf = NeuralForecast(models=models, freq='ME') nf.fit(df=Y_df) ``` > **Tip** > > The performance of Deep Learning models can be very sensitive to the > choice of hyperparameters. Tuning the correct hyperparameters is an > important step to obtain the best forecasts. The `Auto` version of > these models, > [`AutoLSTM`](https://nixtlaverse.nixtla.io/neuralforecast/models.html#autolstm) > and > [`AutoNHITS`](https://nixtlaverse.nixtla.io/neuralforecast/models.html#autonhits), > already perform hyperparameter selection automatically. ### Predict using the fitted models Using the [`NeuralForecast.predict`](https://nixtlaverse.nixtla.io/neuralforecast/core.html#neuralforecast.predict) method you can obtain the `h` forecasts after the training data `Y_df`. ```python theme={null} Y_hat_df = nf.predict() ``` The [`NeuralForecast.predict`](https://nixtlaverse.nixtla.io/neuralforecast/core.html#neuralforecast.predict) method returns a DataFrame with the forecasts for each `unique_id`, `ds`, and model. ```python theme={null} Y_hat_df = Y_hat_df Y_hat_df.head() ``` | | unique\_id | ds | LSTM | NHITS | | - | ---------- | ---------- | ---------- | ---------- | | 0 | 1.0 | 1961-01-31 | 445.602112 | 447.531281 | | 1 | 1.0 | 1961-02-28 | 431.253510 | 439.081024 | | 2 | 1.0 | 1961-03-31 | 456.301270 | 481.924194 | | 3 | 1.0 | 1961-04-30 | 508.149750 | 501.501343 | | 4 | 1.0 | 1961-05-31 | 524.903870 | 514.664551 | ## 4. Plot Predictions Finally, we plot the forecasts of both models againts the real values. ```python theme={null} from utilsforecast.plotting import plot_series ``` ```python theme={null} plot_series(Y_df, Y_hat_df) ``` > **Tip** > > For this guide we are using a simple > [`LSTM`](https://nixtlaverse.nixtla.io/neuralforecast/models.lstm.html#lstm) > model. More recent models, such as > [`TSMixer`](https://nixtlaverse.nixtla.io/neuralforecast/models.tsmixer.html#tsmixer), > [`TFT`](https://nixtlaverse.nixtla.io/neuralforecast/models.tft.html#tft) > and > [`NHITS`](https://nixtlaverse.nixtla.io/neuralforecast/models.nhits.html#nhits) > achieve better accuracy than > [`LSTM`](https://nixtlaverse.nixtla.io/neuralforecast/models.lstm.html#lstm) > in most settings. The full list of available models is available > [here](https://nixtlaverse.nixtla.io/neuralforecast/docs/capabilities/overview.html). ## References * [Boris N. Oreshkin, Dmitri Carpov, Nicolas Chapados, Yoshua Bengio (2020). “N-BEATS: Neural basis expansion analysis for interpretable time series forecasting”. International Conference on Learning Representations.](https://arxiv.org/abs/1905.10437)
* [Cristian Challu, Kin G. Olivares, Boris N. Oreshkin, Federico Garza, Max Mergenthaler-Canseco, Artur Dubrawski (2021). NHITS: Neural Hierarchical Interpolation for Time Series Forecasting. Accepted at AAAI 2023.](https://arxiv.org/abs/2201.12886)