AttributeError: 'Train' object has no attribute 'lambda_delay'

[AmolMavuduru]

I am running NeuralProphet on MacOS Catalina 10.15.5 and executed the following code in a Jupyter Notebook:

model = NeuralProphet(n_forecasts=3,
    n_lags=5,
    ar_sparsity=0.01,
    num_hidden_layers=2,
    yearly_seasonality=False,
    weekly_seasonality=False,
    daily_seasonality=False)

model.fit(sp500_data, 
          freq='D',
          epochs=100)

I got an error titled "AttributeError: 'Train' object has no attribute 'lambda_delay'". Here is the full stack trace:

---

AttributeError Traceback (most recent call last)
in
9 model.fit(sp500_data,
10 freq='D',
---> 11 epochs=100)

~/opt/anaconda3/lib/python3.7/site-packages/neuralprophet/forecaster.py in fit(self, df, freq, epochs, validate_each_epoch, valid_p, use_tqdm, plot_live_loss)
753 metrics_df = self._train(df_train, df_val, use_tqdm=use_tqdm, plot_live_loss=plot_live_loss)
754 else:
--> 755 metrics_df = self._train(df, use_tqdm=use_tqdm, plot_live_loss=plot_live_loss)
756 if epochs is not None:
757 self.config_train.epochs = default_epochs

~/opt/anaconda3/lib/python3.7/site-packages/neuralprophet/forecaster.py in _train(self, df, df_val, use_tqdm, plot_live_loss)
622 if val:
623 val_metrics.reset()
--> 624 epoch_metrics = self._train_epoch(e, loader)
625 metrics_live["{}".format(list(epoch_metrics)[0])] = epoch_metrics[list(epoch_metrics)[0]]
626 if val:

~/opt/anaconda3/lib/python3.7/site-packages/neuralprophet/forecaster.py in _train_epoch(self, e, loader)
483 if self.n_lags > 0: # slowly increase regularization until lambda_delay epoch
484 reg_lambda_ar = utils.get_regularization_lambda(
--> 485 self.config_train.ar_sparsity, self.config_train.lambda_delay, e
486 )
487 for inputs, targets in loader:

AttributeError: 'Train' object has no attribute 'lambda_delay'

[ourownstory]

Hi @AmolMavuduru I am sorry, this was a bug in an earlier version on pip. It should have been resolved in 0.2.7.

To resolve this, you can directly install from github:

git clone <copied link from github>
cd neural_prophet
pip install .

[AmolMavuduru]

Thanks for your response! I installed NeuralProphet directly from GitHub and I see that when I include the number of epochs in my model definition, the code works fine, otherwise, the previous AttributeError occurs again. But for now, the issue is fixed.

[ourownstory]

@AmolMavuduru that should not be the case. Then there is either still a bug, or your venv might be using an older version.
Can you please share if you re-installed it in a new virtual environment, and what version pip shows you? pip list | grep neuralprophet