Model

class flood_forecast.da_rnn.model.MetaMerger(meta_params, meta_method, embed_shape, in_shape)

__init__(meta_params, meta_method, embed_shape, in_shape)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(temporal_data, meta_data)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class flood_forecast.da_rnn.model.DARNN(n_time_series: int, hidden_size_encoder: int, forecast_history: int, decoder_hidden_size: int, out_feats=1, dropout=0.01, meta_data=False, gru_lstm=True, probabilistic=False, final_act=None)

__init__(n_time_series: int, hidden_size_encoder: int, forecast_history: int, decoder_hidden_size: int, out_feats=1, dropout=0.01, meta_data=False, gru_lstm=True, probabilistic=False, final_act=None)

For model benchmark information see link on side https://rb.gy/koozff

Parameters:

• n_time_series (int) – Number of time series present in input

• hidden_size_encoder (int) – dimension of the hidden state encoder

• forecast_history (int) – How many historic time steps to use for forecasting (add one to this number)

• decoder_hidden_size (int) – dimension of hidden size of the decoder

• out_feats (int, optional) – The number of targets (or in classification classes), defaults to 1

• dropout (float, optional) – defaults to .01

• meta_data (bool, optional) – [description], defaults to False

• gru_lstm (bool, optional) – Specify true if you want to use LSTM, defaults to True

• probabilistic (bool, optional) – Specify true if you want to use a probablistic variation, defaults to False

forward(x: Tensor, meta_data: Tensor = None) → Tensor

Performs standard forward pass of the DARNN. Special handling of probablistic.

Parameters:

• x (torch.Tensor) – The core temporal data represented as a tensor (batch_size, forecast_history, n_time_series)

• meta_data (torch( ).Tensor, optional) – The meta-data represented as a tensor (), defaults to None

Returns:

The predictetd number should be in format

Return type:

torch.Tensor

flood_forecast.da_rnn.model.init_hidden(x, hidden_size: int) → Variable

Train the initial value of the hidden state: https://r2rt.com/non-zero-initial-states-for-recurrent-neural-networks.html

class flood_forecast.da_rnn.model.Encoder(input_size: int, hidden_size: int, T: int, gru_lstm: bool = True, meta_data: bool = False)

__init__(input_size: int, hidden_size: int, T: int, gru_lstm: bool = True, meta_data: bool = False)

input size: number of underlying factors (81) T: number of time steps (10) hidden_size: dimension of the hidden stats

forward(input_data: Tensor, meta_data=None) → Tuple[Tensor, Tensor]

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class flood_forecast.da_rnn.model.Decoder(encoder_hidden_size: int, decoder_hidden_size: int, T: int, out_feats=1, gru_lstm: bool = True, probabilistic: bool = True)

__init__(encoder_hidden_size: int, decoder_hidden_size: int, T: int, out_feats=1, gru_lstm: bool = True, probabilistic: bool = True)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(input_encoded: Tensor, y_history: Tensor) → Tensor

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.