https://github.com/nixtla logo
#neural-forecast
Title
# neural-forecast
r

Raul Gherman

09/03/2023, 11:34 AM
HI <!channel>, I am trying to use AutoLSTM with BayesOptSearch() and I don't know what ray tune function to use in config. I tried to use tune.randn and had this error: "OverflowError: Range exceeds valid bounds" and i can't solve it even if I use a very small SD, and I tried also tune.uniform and had this error: "TypeError: 'numpy.float64' object cannot be interpreted as an integer". Does anyone know how to solve this?
c

Cristian (Nixtla)

09/04/2023, 3:09 PM
Hi @Raul Gherman! For what hyperparameter are you trying to use tune.randn? Can you show your line of code please?
r

Raul Gherman

09/06/2023, 8:08 AM
Copy code
LSTMConfing = {
    "encoder_hidden_size": tune.randint(10, 190),
    "encoder_n_layers": tune.randint(1, 4),
    "context_size": tune.randint(5, 20),
    "decoder_hidden_size": tune.randint(16, 256),
    "learning_rate": tune.uniform(1e-4, 1e-2),
    "max_steps": tune.randint(50, 250),
    "batch_size": tune.randint(16, 32),
    "input_size": 16,
    "hist_exog_list": [
        "FedInterest",
        "High",
        "Low",
        "Open",
        "SMA_1M",
        "SMA_1W",
        "SMA_3M",
        "SMA_V_1M",
        "SMA_V_1W",
        "SMA_V_3M",
        "Volume",
        "Week Number",
        "Week Number Average",
    ],

}
c

Cristian (Nixtla)

09/06/2023, 1:48 PM
The config seems correct. Can you show the complete output of the error for "OverflowError: Range exceeds valid bounds"? To see which part of the architecture is causing the issue. tune.uniform will produce float values, so you cant use it for int only hyperparameters such as
context_size
or number of layers.
2 Views