feat!: old vanilla AE

[ab93]

Summary of changes:

• Change the order of postprocess operations; compute feature aggregation first, before postprocess
• Introduce a robust scaling mechanism to lessen the effect of outliers in training

Config to be deployed:

model:
  name: LSTMAE
  conf:
    seq_len: 10
    no_features: 2
    embedding_dim: 2
    loss_fn: "mse"
    weight_decay: 1e-4
preprocess:
  - name: LogTransformer
    conf:
      add_factor: 2
  - name: PercentileScaler
    conf:
        max_percentile: 90
threshold:
  name: MaxPercentileThreshold
  conf:
    min_threshold: 0.1
trainer:
  batch_size: 256
  pltrainer_conf:
    accelerator: cpu
    max_epochs: 40
    deterministic: true
score:
  window_agg:
    method: EXP
    conf:
        beta: 0.85
  feature_agg:
    method: WEIGHTED_AVG
    conf:
        weights: [ 0.7, 0.3 ]

Image: quay.io/numaio/numalogic/udf:v0.9.1a0

[codecov]

Codecov Report

Attention: Patch coverage is 74.17840% with 55 lines in your changes are missing coverage. Please review.

Project coverage is 91.95%. Comparing base (f29f771) to head (b20f343).

Files	Patch %	Lines
numalogic/transforms/_scaler.py	33.33%	24 Missing ⚠️
numalogic/connectors/druid/_druid.py	53.33%	12 Missing and 2 partials ⚠️
numalogic/models/threshold/_median.py	45.45%	5 Missing and 1 partial ⚠️
numalogic/transforms/_postprocess.py	42.85%	4 Missing ⚠️
numalogic/backtest/_prom.py	82.35%	1 Missing and 2 partials ⚠️
numalogic/models/autoencoder/variants/vanilla.py	94.87%	0 Missing and 2 partials ⚠️
numalogic/udfs/postprocess.py	83.33%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #368      +/-   ##
==========================================
- Coverage   93.07%   91.95%   -1.13%     
==========================================
  Files          97       98       +1     
  Lines        4492     4612     +120     
  Branches      387      409      +22     
==========================================
+ Hits         4181     4241      +60     
- Misses        231      281      +50     
- Partials       80       90      +10     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ab93]

Closing