hpipy.extensions.NeuralNetworkIndex#

class hpipy.extensions.NeuralNetworkIndex(**kwargs)[source]

Bases: HedonicIndex

Neural network house price index.

Estimates the relationship between independent variables and the dependent variable using a neural network model. The index can be derived from the model using two distinct approaches:

  1. Residual Approach: This method extracts the house price index directly from the market pathway of the neural network. It isolates the temporal component of price changes by zeroing out all non-time features and examining the network’s output, effectively capturing the “residual” market trend.

  2. Attributional Approach: This method derives the index by analyzing the explainability of both market and time components. It uses attribution techniques to decompose the network’s predictions and to assess the conditional marginal impact of each time period on price changes, which are converted into an index.

Both approaches aim to capture market trends, but differ in how they extract temporal information from the neural network’s learned representations.

Methods

__init__

Initialize base house price index.

coef_to_index

Convert coefficients to an index.

create_index

Create the index.

create_series

Create a series from the index.

from_model

Create an index from a house price model.

get_data

Get hedonic transaction data.

get_model

Get a neural network house price model.

smooth_index

Smooth the index.

smooth_series

Apply smoothing to all indices.

Attributes

data

model

name

periods

value

index

imputed

smooth

accuracy

accuracy_smooth

volatility

volatility_smooth

revision

revision_smooth

Parameters:

kwargs (Any)

static get_model()[source]

Get a neural network house price model.

Return type:

type[BaseHousePriceModel]

__init__(**kwargs)

Initialize base house price index.

Parameters:

kwargs (Any)

Return type:

None

static coef_to_index(coef_df, log_dep, base_price=1)

Convert coefficients to an index.

Parameters:

  • coef_df (pd.DataFrame) – Coefficients.

  • log_dep (bool) – Log dependent variable.

  • base_price (float, optional) – Base price. Defaults to 1.

Returns:

Index and imputed values.

Return type:

tuple[np.ndarray, np.ndarray]

classmethod create_index(trans_data, prop_id=None, trans_id=None, price=None, seq_only=True, max_period=None, smooth=False, periodicity=None, min_date=None, max_date=None, adj_type=None, **kwargs)

Create the index.

Parameters:

  • trans_data (TransactionData | PeriodTable | pd.DataFrame) – Input transaction data.

  • prop_id (str | None, optional) – Property identifier. Defaults to None.

  • trans_id (str | None, optional) – Transaction identifier. Defaults to None.

  • price (str | None, optional) – Price column name. Defaults to None.

  • seq_only (bool, optional) – Sequential only. Defaults to True.

  • max_period (int | None, optional) – Maximum index period. Defaults to None.

  • smooth (bool, optional) – Smooth the index. Defaults to False.

  • periodicity (str | None, optional) – Periodicity of the index. Defaults to None.

  • min_date (str | None, optional) – Minimum date for the index. Defaults to None.

  • max_date (str | None, optional) – Maximum date for the index. Defaults to None.

  • adj_type (str | None, optional) – Adjustment type. Defaults to None.

  • **kwargs – Additional keyword arguments.

Returns:

Index object.

Return type:

Self

create_series(train_period=12, max_period=None, **kwargs)

Create a series from the index.

Parameters:

  • train_period (int, optional) – Train period for the series. Defaults to 12.

  • max_period (int | None, optional) – Maximum period for the series. Defaults to None.

  • **kwargs – Additional keyword arguments.

Returns:

Index object.

Return type:

Self

classmethod from_model(model, trans_data=None, max_period=None, smooth=False, **kwargs)

Create an index from a house price model.

Parameters:

  • model (BaseHousePriceModel) – House price model object.

  • trans_data (TransactionData | None, optional) – Transaction data object. Defaults to None.

  • max_period (int | None, optional) – Maximum period for the index. Defaults to None.

  • smooth (bool, optional) – Smooth the index. Defaults to False.

  • **kwargs – Additional keyword arguments.

Returns:

Index object.

Return type:

Self

static get_data()

Get hedonic transaction data.

Return type:

type[TransactionData]

smooth_index(order=3, in_place=False)

Smooth the index.

Parameters:

  • order (list[int] | int, optional) – Smoothing order. Defaults to 3.

  • in_place (bool, optional) – Smooth in place or return new object. Defaults to False.

Returns:

Smoothed index or new smoothed index object.

Return type:

Self | pd.Series

smooth_series(order=3)

Apply smoothing to all indices.

Parameters:

order (int) – Defaults to 3.

Returns:

Index object.

Return type:

Self

data: TransactionData
model: BaseHousePriceModel
name: pd.Series
periods: pd.Series
value: pd.Series
index: Any
imputed: np.ndarray
smooth: Any
accuracy: pd.DataFrame
accuracy_smooth: pd.DataFrame
volatility: pd.DataFrame
volatility_smooth: pd.DataFrame
revision: pd.DataFrame
revision_smooth: pd.DataFrame

This Page