fix: scaler documentation

online-ml · Jul 7, 2022 · d80a8ec · d80a8ec
1 parent c82d3a4
commit d80a8ec
Show file tree

Hide file tree

Showing 8 changed files with 116 additions and 23 deletions.
diff --git a/README.md b/README.md
@@ -14,25 +14,31 @@
 </p>
 
 ## 💈 Installation
+
 ```shell
 pip install river-torch
 ```
+
 You can install the latest development version from GitHub as so:
+
 ```shell
 pip install https://github.com/online-ml/river-torch --upgrade
 ```
 
 Or, through SSH:
+
 ```shell
 pip install git@github.com:online-ml/river-torch.git --upgrade
 ```
 
-
 ## 🍫 Quickstart
+
 We build the development of neural networks on top of the <a href="https://www.riverml.xyz">river API</a> and refer to the rivers design principles.
 The following example creates a simple MLP architecture based on PyTorch and incrementally predicts and trains on the website phishing dataset.
 For further examples check out the <a href="https://online-ml.github.io/river-torch">Documentation</a>.
+
 ### Classification
+
 ```python
 >>> from river import datasets
 >>> from river import metrics
@@ -81,7 +87,7 @@ Accuracy: 0.8304
 ```python
 >>> import math
 >>> from river import datasets, metrics
->>> from river_torch.anomaly import AutoEncoder
+>>> from river_torch.anomaly import Autoencoder
 >>> from river_torch.utils import get_activation_fn
 >>> from torch import manual_seed, nn
 
@@ -113,7 +119,7 @@ Accuracy: 0.8304
 >>> encoder_fn = get_encoder
 >>> decoder_fn = get_decoder
 
->>> model = AutoEncoder(encoder_fn=encoder_fn,decoder_fn=decoder_fn, lr=0.01)
+>>> model = Autoencoder(encoder_fn=encoder_fn,decoder_fn=decoder_fn, lr=0.01)
 
 >>> for x,y in dataset:
 ...     score = model.score_one(x)
@@ -123,6 +129,7 @@ Accuracy: 0.8304
 ```
 
 ## 🏫 Affiliations
+
 <p align="center">
     <img src="https://upload.wikimedia.org/wikipedia/de/thumb/4/44/Fzi_logo.svg/1200px-Fzi_logo.svg.png?raw=true" alt="FZI Logo" height="200"/>
 </p>
diff --git a/docs/examples/example_ae.ipynb b/docs/examples/example_ae.ipynb
@@ -17,9 +17,9 @@
      "traceback": [
       "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
       "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
-      "\u001b[1;32m/home/lucascazzonelli/Documents/Research/river-torch/docs/examples/example_ae.ipynb Cell 1'\u001b[0m in \u001b[0;36m<cell line: 2>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      <a href='vscode-notebook-cell:/home/lucascazzonelli/Documents/Research/river-torch/docs/examples/example_ae.ipynb#ch0000000?line=0'>1</a>\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mriver\u001b[39;00m \u001b[39mimport\u001b[39;00m datasets, metrics\n\u001b[0;32m----> <a href='vscode-notebook-cell:/home/lucascazzonelli/Documents/Research/river-torch/docs/examples/example_ae.ipynb#ch0000000?line=1'>2</a>\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mriver_torch\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39manomaly\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39mnn_builder\u001b[39;00m \u001b[39mimport\u001b[39;00m get_fc_encoder, get_fc_decoder\n\u001b[1;32m      <a href='vscode-notebook-cell:/home/lucascazzonelli/Documents/Research/river-torch/docs/examples/example_ae.ipynb#ch0000000?line=2'>3</a>\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mriver_torch\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39manomaly\u001b[39;00m \u001b[39mimport\u001b[39;00m AutoEncoder\n",
+      "\u001b[1;32m/home/lucascazzonelli/Documents/Research/river-torch/docs/examples/example_ae.ipynb Cell 1'\u001b[0m in \u001b[0;36m<cell line: 2>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      <a href='vscode-notebook-cell:/home/lucascazzonelli/Documents/Research/river-torch/docs/examples/example_ae.ipynb#ch0000000?line=0'>1</a>\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mriver\u001b[39;00m \u001b[39mimport\u001b[39;00m datasets, metrics\n\u001b[0;32m----> <a href='vscode-notebook-cell:/home/lucascazzonelli/Documents/Research/river-torch/docs/examples/example_ae.ipynb#ch0000000?line=1'>2</a>\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mriver_torch\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39manomaly\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39mnn_builder\u001b[39;00m \u001b[39mimport\u001b[39;00m get_fc_encoder, get_fc_decoder\n\u001b[1;32m      <a href='vscode-notebook-cell:/home/lucascazzonelli/Documents/Research/river-torch/docs/examples/example_ae.ipynb#ch0000000?line=2'>3</a>\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mriver_torch\u001b[39;00m\u001b[39m.\u001b[39;00m\u001b[39manomaly\u001b[39;00m \u001b[39mimport\u001b[39;00m Autoencoder\n",
       "File \u001b[0;32m~/Documents/Research/river-torch/river_torch/__init__.py:1\u001b[0m, in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39m.\u001b[39;00m \u001b[39mimport\u001b[39;00m anomaly, classification, regression, utils\n\u001b[1;32m      2\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39m.\u001b[39;00m\u001b[39m__version__\u001b[39;00m \u001b[39mimport\u001b[39;00m __version__  \u001b[39m# noqa: F401\u001b[39;00m\n\u001b[1;32m      4\u001b[0m __all__ \u001b[39m=\u001b[39m [\u001b[39m\"\u001b[39m\u001b[39manomaly\u001b[39m\u001b[39m\"\u001b[39m, \u001b[39m\"\u001b[39m\u001b[39mclassification\u001b[39m\u001b[39m\"\u001b[39m, \u001b[39m\"\u001b[39m\u001b[39mregression\u001b[39m\u001b[39m\"\u001b[39m, \u001b[39m\"\u001b[39m\u001b[39mutils\u001b[39m\u001b[39m\"\u001b[39m]\n",
-      "File \u001b[0;32m~/Documents/Research/river-torch/river_torch/anomaly/__init__.py:7\u001b[0m, in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39m.\u001b[39;00m\u001b[39mrolling_ae\u001b[39;00m \u001b[39mimport\u001b[39;00m RollingWindowAutoencoder\n\u001b[1;32m      6\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39m.\u001b[39;00m\u001b[39mbase\u001b[39;00m \u001b[39mimport\u001b[39;00m AutoEncoder\n\u001b[0;32m----> 7\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39m.\u001b[39;00m\u001b[39mscaler\u001b[39;00m \u001b[39mimport\u001b[39;00m (\n\u001b[1;32m      8\u001b[0m     StandardScaler,\n\u001b[1;32m      9\u001b[0m     MeanScaler,\n\u001b[1;32m     10\u001b[0m     MinMaxScaler,\n\u001b[1;32m     11\u001b[0m     RollingStandardScaler,\n\u001b[1;32m     12\u001b[0m     AdaptiveStandardScaler,\n\u001b[1;32m     13\u001b[0m     RollingMinMaxScaler,\n\u001b[1;32m     14\u001b[0m     RollingMeanScaler,\n\u001b[1;32m     15\u001b[0m     AdaptiveMeanScaler,\n\u001b[1;32m     16\u001b[0m )\n\u001b[1;32m     18\u001b[0m __all__ \u001b[39m=\u001b[39m [\n\u001b[1;32m     19\u001b[0m     \u001b[39m\"\u001b[39m\u001b[39mAutoEncoder\u001b[39m\u001b[39m\"\u001b[39m,\n\u001b[1;32m     20\u001b[0m     \u001b[39m\"\u001b[39m\u001b[39mVariationalAutoencoder\u001b[39m\u001b[39m\"\u001b[39m,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m     31\u001b[0m     \u001b[39m\"\u001b[39m\u001b[39mAdaptiveMeanScaler\u001b[39m\u001b[39m\"\u001b[39m,\n\u001b[1;32m     32\u001b[0m ]\n",
+      "File \u001b[0;32m~/Documents/Research/river-torch/river_torch/anomaly/__init__.py:7\u001b[0m, in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39m.\u001b[39;00m\u001b[39mrolling_ae\u001b[39;00m \u001b[39mimport\u001b[39;00m RollingWindowAutoencoder\n\u001b[1;32m      6\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39m.\u001b[39;00m\u001b[39mbase\u001b[39;00m \u001b[39mimport\u001b[39;00m Autoencoder\n\u001b[0;32m----> 7\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39m.\u001b[39;00m\u001b[39mscaler\u001b[39;00m \u001b[39mimport\u001b[39;00m (\n\u001b[1;32m      8\u001b[0m     StandardScaler,\n\u001b[1;32m      9\u001b[0m     MeanScaler,\n\u001b[1;32m     10\u001b[0m     MinMaxScaler,\n\u001b[1;32m     11\u001b[0m     RollingStandardScaler,\n\u001b[1;32m     12\u001b[0m     AdaptiveStandardScaler,\n\u001b[1;32m     13\u001b[0m     RollingMinMaxScaler,\n\u001b[1;32m     14\u001b[0m     RollingMeanScaler,\n\u001b[1;32m     15\u001b[0m     AdaptiveMeanScaler,\n\u001b[1;32m     16\u001b[0m )\n\u001b[1;32m     18\u001b[0m __all__ \u001b[39m=\u001b[39m [\n\u001b[1;32m     19\u001b[0m     \u001b[39m\"\u001b[39m\u001b[39mAutoencoder\u001b[39m\u001b[39m\"\u001b[39m,\n\u001b[1;32m     20\u001b[0m     \u001b[39m\"\u001b[39m\u001b[39mVariationalAutoencoder\u001b[39m\u001b[39m\"\u001b[39m,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m     31\u001b[0m     \u001b[39m\"\u001b[39m\u001b[39mAdaptiveMeanScaler\u001b[39m\u001b[39m\"\u001b[39m,\n\u001b[1;32m     32\u001b[0m ]\n",
       "File \u001b[0;32m~/Documents/Research/river-torch/river_torch/anomaly/scaler.py:105\u001b[0m, in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m    100\u001b[0m         score \u001b[39m=\u001b[39m raw_score \u001b[39m/\u001b[39m mean\n\u001b[1;32m    102\u001b[0m         \u001b[39mreturn\u001b[39;00m score\n\u001b[0;32m--> 105\u001b[0m \u001b[39mclass\u001b[39;00m \u001b[39mMinMaxScaler\u001b[39;00m(AnomalyScaler):\n\u001b[1;32m    106\u001b[0m     \u001b[39mdef\u001b[39;00m \u001b[39m__init__\u001b[39m(\u001b[39mself\u001b[39m, anomaly_detector: anomaly\u001b[39m.\u001b[39mAnomalyDetector):\n\u001b[1;32m    107\u001b[0m         \u001b[39msuper\u001b[39m()\u001b[39m.\u001b[39m\u001b[39m__init__\u001b[39m(anomaly_detector)\n",
       "File \u001b[0;32m~/Documents/Research/river-torch/river_torch/anomaly/scaler.py:106\u001b[0m, in \u001b[0;36mMinMaxScaler\u001b[0;34m()\u001b[0m\n\u001b[1;32m    105\u001b[0m \u001b[39mclass\u001b[39;00m \u001b[39mMinMaxScaler\u001b[39;00m(AnomalyScaler):\n\u001b[0;32m--> 106\u001b[0m     \u001b[39mdef\u001b[39;00m \u001b[39m__init__\u001b[39m(\u001b[39mself\u001b[39m, anomaly_detector: anomaly\u001b[39m.\u001b[39;49mAnomalyDetector):\n\u001b[1;32m    107\u001b[0m         \u001b[39msuper\u001b[39m()\u001b[39m.\u001b[39m\u001b[39m__init__\u001b[39m(anomaly_detector)\n\u001b[1;32m    108\u001b[0m         \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mmin \u001b[39m=\u001b[39m Min()\n",
       "\u001b[0;31mAttributeError\u001b[0m: partially initialized module 'river_torch.anomaly' has no attribute 'AnomalyDetector' (most likely due to a circular import)"
@@ -29,7 +29,7 @@
    "source": [
     "from river import datasets, metrics\n",
     "from river_torch.anomaly.nn_builder import get_fc_encoder, get_fc_decoder\n",
-    "from river_torch.anomaly import AutoEncoder"
+    "from river_torch.anomaly import Autoencoder"
    ]
   },
   {
@@ -62,7 +62,7 @@
     "dataset = datasets.CreditCard().take(5000)\n",
     "metric = metrics.ROCAUC()\n",
     "\n",
-    "model = AutoEncoder(encoder_fn=encoder_fn,decoder_fn=decoder_fn, lr=0.01, n_features=5)\n",
+    "model = Autoencoder(encoder_fn=encoder_fn,decoder_fn=decoder_fn, lr=0.01, n_features=5)\n",
     "\n",
     "for x,y in dataset:\n",
     "    score = model.score_one(x)\n",

diff --git a/river_torch/anomaly/__init__.py b/river_torch/anomaly/__init__.py
@@ -1,9 +1,9 @@
-from .base import AutoEncoder
+from .base import Autoencoder
 from .probability_weighted_ae import ProbabilityWeightedAutoencoder
 from .variational_ae import VariationalAutoencoder
 from .rolling_ae import RollingWindowAutoencoder
 
-from .base import AutoEncoder, AnomalyScaler
+from .base import Autoencoder, AnomalyScaler
 from .scaler import (
     StandardScaler,
     MeanScaler,
@@ -16,7 +16,7 @@
 )
 
 __all__ = [
-    "AutoEncoder",
+    "Autoencoder",
     "VariationalAutoencoder",
     "RollingWindowAutoencoder",
     "WindowedStandardizer",

diff --git a/river_torch/anomaly/base.py b/river_torch/anomaly/base.py
@@ -11,7 +11,7 @@
 from river_torch.utils import dict2tensor, get_loss_fn, get_optimizer_fn
 
 
-class AutoEncoder(anomaly.AnomalyDetector, nn.Module):
+class Autoencoder(anomaly.AnomalyDetector, nn.Module):
     """
     Base Auto Encoder
     ----------
@@ -182,9 +182,35 @@ def configure_optimizers(self):
 
 
 class AnomalyScaler(base.Wrapper, anomaly.AnomalyDetector):
+    """AnomalyScaler is a wrapper around an anomaly detector that scales the output of the model.
+
+    Parameters
+    ----------
+    anomaly_detector
+    """
+
     def __init__(self, anomaly_detector: anomaly.AnomalyDetector):
         self.anomaly_detector = anomaly_detector
 
+    @classmethod
+    def _unit_test_params(self):
+        yield {"anomaly_detector": anomaly.HalfSpaceTrees()}
+
+    @classmethod
+    def _unit_test_skips(self):
+        """Indicates which checks to skip during unit testing.
+        Most estimators pass the full test suite. However, in some cases, some estimators might not
+        be able to pass certain checks.
+        """
+        return {
+            "check_pickling",
+            "check_shuffle_features_no_impact",
+            "check_emerging_features",
+            "check_disappearing_features",
+            "check_predict_proba_one",
+            "check_predict_proba_one_binary",
+        }
+
     @property
     def _wrapped_model(self):
         return self.anomaly_detector

diff --git a/river_torch/anomaly/probability_weighted_ae.py b/river_torch/anomaly/probability_weighted_ae.py
@@ -9,7 +9,8 @@
 from river_torch.anomaly import base
 from river_torch.utils import dict2tensor
 
-class ProbabilityWeightedAutoencoder(base.AutoEncoder):
+
+class ProbabilityWeightedAutoencoder(base.Autoencoder):
     """
     A propability weighted auto encoder
     ----------
@@ -72,4 +73,4 @@ def learn_one(self, x):
         self.optimizer.zero_grad()
         loss.backward()
         self.optimizer.step()
-        return self
+        return self
diff --git a/river_torch/anomaly/rolling_ae.py b/river_torch/anomaly/rolling_ae.py
@@ -4,7 +4,8 @@
 from river_torch.anomaly import base
 from river_torch.utils import dict2tensor
 
-class RollingWindowAutoencoder(base.AutoEncoder):
+
+class RollingWindowAutoencoder(base.Autoencoder):
     """
     A rolling window auto encoder
     ----------
@@ -62,4 +63,4 @@ def learn_one(self, x):
         if len(self._x_window) == self.window_size:
             x = torch.concat(list(self._x_window.values))
             self._learn_batch(x=x)
-        return self
+        return self
diff --git a/river_torch/anomaly/scaler.py b/river_torch/anomaly/scaler.py
@@ -3,14 +3,22 @@
 
 
 class StandardScaler(AnomalyScaler):
+    """Wrapper around an anomaly detector that standardizes the model's output using incremental mean and variance metrics.
+
+    Parameters
+    ----------
+    anomaly_detector
+    with_std : bool
+    """
+
     def __init__(self, anomaly_detector, with_std=True):
         super().__init__(anomaly_detector)
         self.mean = Mean()
         self.sq_mean = Mean()
         self.with_std = with_std
 
     def score_one(self, *args) -> float:
-        """Return calibrated anomaly score based on raw score provided by the wrapped anomaly detector.
+        """Return scaled anomaly score based on raw score provided by the wrapped anomaly detector.
 
         A high score is indicative of an anomaly. A low score corresponds to a normal observation.
         Parameters
@@ -33,7 +41,7 @@ def score_one(self, *args) -> float:
         return score
 
     def score_many(self, *args):
-        """Return calibrated anomaly scores based on raw scores provided by the wrapped anomaly detector.
+        """Return scaled anomaly scores based on raw scores provided by the wrapped anomaly detector.
 
         A high score is indicative of an anomaly. A low score corresponds to a normal observation.
         Parameters
@@ -57,12 +65,18 @@ def score_many(self, *args):
 
 
 class MeanScaler(AnomalyScaler):
+    """ Wrapper around an anomaly detector that scales the model's output by the incremental mean of previous scores.
+
+    Parameters
+    ----------
+    anomaly_detector
+    """
     def __init__(self, anomaly_detector):
         super().__init__(anomaly_detector=anomaly_detector)
         self.mean = Mean()
 
     def score_one(self, *args) -> float:
-        """Return calibrated anomaly score based on raw score provided by the wrapped anomaly detector.
+        """Return scaled anomaly score based on raw score provided by the wrapped anomaly detector.
 
         A high score is indicative of an anomaly. A low score corresponds to a normal observation.
         Parameters
@@ -81,7 +95,7 @@ def score_one(self, *args) -> float:
         return score
 
     def score_many(self, *args) -> float:
-        """Return calibrated anomaly scores based on raw scores provided by the wrapped anomaly detector.
+        """Return scaled anomaly scores based on raw scores provided by the wrapped anomaly detector.
 
         A high score is indicative of an anomaly. A low score corresponds to a normal observation.
         Parameters
@@ -101,13 +115,19 @@ def score_many(self, *args) -> float:
 
 
 class MinMaxScaler(AnomalyScaler):
+    """Wrapper around an anomaly detector that scales the model's output to $[0, 1]$ using rolling min and max metrics.
+
+    Parameters
+    ----------
+    anomaly_detector
+    """
     def __init__(self, anomaly_detector):
         super().__init__(anomaly_detector)
         self.min = Min()
         self.max = Max()
 
     def score_one(self, *args) -> float:
-        """Return calibrated anomaly score based on raw score provided by the wrapped anomaly detector.
+        """Return scaled anomaly score based on raw score provided by the wrapped anomaly detector.
 
         A high score is indicative of an anomaly. A low score corresponds to a normal observation.
         Parameters
@@ -127,7 +147,7 @@ def score_one(self, *args) -> float:
         return score
 
     def score_many(self, *args) -> float:
-        """Return calibrated anomaly score based on raw score provided by the wrapped anomaly detector.
+        """Return scaled anomaly score based on raw score provided by the wrapped anomaly detector.
 
         A high score is indicative of an anomaly. A low score corresponds to a normal observation.
         Parameters
@@ -152,6 +172,15 @@ def score_many(self, *args) -> float:
 
 
 class RollingStandardScaler(StandardScaler):
+    """Wrapper around an anomaly detector that standardizes the model's output using rolling mean and variance metrics.
+
+    Parameters
+    ----------
+    anomaly_detector
+    window_size
+    with_std : bool
+    """
+
     def __init__(self, anomaly_detector, window_size=250, with_std=True):
         super().__init__(anomaly_detector=anomaly_detector)
         self.window_size = window_size
@@ -161,15 +190,30 @@ def __init__(self, anomaly_detector, window_size=250, with_std=True):
 
 
 class AdaptiveStandardScaler(StandardScaler):
+    """Wrapper around an anomaly detector that standardizes the model's output using exponential running mean and variance metrics.
+
+    Parameters
+    ----------
+    anomaly_detector
+    alpha
+    with_std
+    """
     def __init__(self, anomaly_detector, alpha=0.3, with_std=True):
         super().__init__(anomaly_detector=anomaly_detector)
-        self.apha = alpha
+        self.alpha = alpha
         self.mean = EWMean(alpha=alpha)
         self.sq_mean = EWMean(alpha=alpha) if with_std else None
         self.with_std = with_std
 
 
 class RollingMinMaxScaler(MinMaxScaler):
+    """Wrapper around an anomaly detector that scales the model's output to $[0, 1]$ using rolling min and max metrics.
+
+    Parameters
+    ----------
+    anomaly_detector
+    window_size
+    """
     def __init__(self, anomaly_detector, window_size=250):
         super().__init__(anomaly_detector=anomaly_detector)
         self.window_size = window_size
@@ -178,13 +222,27 @@ def __init__(self, anomaly_detector, window_size=250):
 
 
 class RollingMeanScaler(MeanScaler):
+    """ Wrapper around an anomaly detector that scales the model's output by the rolling mean of previous scores.
+
+    Parameters
+    ----------
+    anomaly_detector
+    window_size
+    """
     def __init__(self, anomaly_detector, window_size=250):
         super().__init__(anomaly_detector)
         self.window_size = window_size
         self.mean = RollingMean(window_size=window_size)
 
 
 class AdaptiveMeanScaler(MeanScaler):
+    """ Wrapper around an anomaly detector that scales the model's output by the exponential running mean of previous scores.
+
+    Parameters
+    ----------
+    anomaly_detector
+    alpha
+    """
     def __init__(self, anomaly_detector, alpha=0.3):
         super().__init__(anomaly_detector)
         self.alpha = alpha

diff --git a/river_torch/anomaly/variational_ae.py b/river_torch/anomaly/variational_ae.py
@@ -6,7 +6,7 @@
 from river_torch.utils import dict2tensor
 
 
-class VariationalAutoencoder(base.AutoEncoder):
+class VariationalAutoencoder(base.Autoencoder):
     """
     A propability weighted auto encoder
     ----------