learn scores in the orignal scale

examples/anomaly_detection_pytorch.py

@@ -126,14 +126,14 @@ def main(args):
  )
  # remove the very last param from the params
  sliced_params = [p[:, :-1] for p in params]
- distr = model.output_distribution(sliced_params)
+ distr = model.output_distribution(sliced_params, scale=scale)
 
  # get the last target and calcualte its anomaly score
  context_target = take_last(
  inputs["past_target"], dim=-1, num=model.context_length - 1
  )
  # calculate the surprisal scores for the context target
- scores = -distr.log_prob(context_target / scale)
+ scores = -distr.log_prob(context_target)
 
  # get the args.top_score_percentage of the scores for each time series of the batch
  top_scores = torch.topk(
@@ -149,15 +149,18 @@ def main(args):
  )
 
  # Loop over each prediction length
- scaled_future_target = inputs["future_target"] / scale
- distr = model.output_distribution(params, trailing_n=1)
+ distr = model.output_distribution(
+ params, trailing_n=1, scale=scale
+ )
  batch_anomalies = []
  for i in tqdm(
- range(scaled_future_target.shape[1]),
+ range(inputs["future_target"].shape[1]),
  desc="Processing prediction length",
  leave=False,
  ):
- score = -distr.log_prob(scaled_future_target[:, i : i + 1])
+ target = inputs["future_target"][:, i : i + 1]
+ score = -distr.log_prob(target)
+
  # only check if its an anomaly for scores greater than gpd.loc for each entry in the batch
  is_anomaly = torch.where(
  score < gpd.loc,
@@ -173,19 +176,20 @@ def main(args):
  ),
  dim=-1,
  )
-
  next_lags = lagged_sequence_values(
  model.lags_seq,
  inputs["past_target"] / scale,
- scaled_future_target[:, i : i + 1],
+ target / scale,
  dim=-1,
  )
  rnn_input = torch.cat((next_lags, next_features), dim=-1)
+
  output, state = model.rnn(rnn_input, state)
  params = model.param_proj(output)
- distr = model.output_distribution(params)
+ distr = model.output_distribution(params, scale=scale)
  # stack the batch_anomalies along the prediction length dimension
  anomalies.append(torch.stack(batch_anomalies, dim=1))
+ # concat the anomalies along the batch dimension
  anomalies = torch.cat(anomalies, dim=0).cpu().numpy()
 
  # save as csv
@@ -238,7 +242,7 @@ def main(args):
  parser.add_argument(
  "--top_score_percentage",
  type=float,
- default=0.1,
+ default=0.2,
  help="Percentage of top scores to consider for GPD fitting",
  )
  parser.add_argument(