GITBOOK-8: No subject

Author: Rajakavitha1

docs-gb/README.md

@@ -1,16 +1,13 @@
+# Alibi Detect
+
 ![Alibi Detect Logo](images/Alibi_Detect_Logo_rgb.png)
 
-# Alibi Detect
+## Alibi Detect
 
 [Alibi Detect](https://github.com/SeldonIO/alibi-detect) is a source-available Python library focused on **outlier**, **adversarial** and **drift** detection. The package aims to cover both online and offline detectors for tabular data, text, images and time series. Both **TensorFlow** and **PyTorch** backends are supported for drift detection.
 
-For more background on the importance of monitoring outliers and distributions in a production setting, check out 
-[this talk](https://slideslive.com/38931758/monitoring-and-explainability-of-models-in-production?ref=speaker-37384-latest) 
-from the *Challenges in Deploying and Monitoring Machine Learning Systems* ICML 2020 workshop, based on the paper 
-[Monitoring and explainability of models in production](https://arxiv.org/abs/2007.06299) and referencing Alibi Detect.
+For more background on the importance of monitoring outliers and distributions in a production setting, check out [this talk](https://slideslive.com/38931758/monitoring-and-explainability-of-models-in-production?ref=speaker-37384-latest) from the _Challenges in Deploying and Monitoring Machine Learning Systems_ ICML 2020 workshop, based on the paper [Monitoring and explainability of models in production](https://arxiv.org/abs/2007.06299) and referencing Alibi Detect.
 
-For a thorough introduction to drift detection, check out the talk below titled, [Protecting Your Machine Learning Against Drift: An Introduction](https://youtu.be/tL5sEaQha5o). 
-The talk covers what drift is and why it pays to detect it, the different types of drift, how it 
-can be detected in a principled manner and also describes the anatomy of a drift detector.
+For a thorough introduction to drift detection, check out the talk below titled, [Protecting Your Machine Learning Against Drift: An Introduction](https://youtu.be/tL5sEaQha5o). The talk covers what drift is and why it pays to detect it, the different types of drift, how it can be detected in a principled manner and also describes the anatomy of a drift detector.
 
 {% embed url="https://youtu.be/tL5sEaQha5o" %}

docs-gb/SUMMARY.md

@@ -1,11 +1,10 @@
 # Table of contents
 
 * [Alibi Detect](README.md)
-* [Getting Started](overview/getting\_started.md)
+* [Getting Started](overview/getting_started.md)
 * [Algorithm Overview](overview/algorithms.md)
 * [Saving and Loading](overview/saving.md)
-* [Detector Configuration Files](overview/config\_files.md)
-* [Roadmap](overview/roadmap.md)
+* [Detector Configuration Files](overview/config_files.md)
 * [Outlier Detection](outlier-detection/README.md)
   * [Methods](outlier-detection/methods/README.md)
     * [Mahalanobis Distance](od/methods/mahalanobis.md)
@@ -19,19 +18,19 @@
     * [Spectral Residual](od/methods/sr.md)
     * [Sequence-to-Sequence (Seq2Seq)](od/methods/seq2seq.md)
   * [Examples](outlier-detection/examples/README.md)
-    * [AE outlier detection on CIFAR10](examples/od\_ae\_cifar10.md)
-    * [AEGMM and VAEGMM outlier detection on KDD Cup ‘99 dataset](examples/od\_aegmm\_kddcup.md)
-    * [Isolation Forest outlier detection on KDD Cup ‘99 dataset](examples/od\_if\_kddcup.md)
-    * [Likelihood Ratio Outlier Detection on Genomic Sequences](examples/od\_llr\_genome.md)
-    * [Likelihood Ratio Outlier Detection with PixelCNN++](examples/od\_llr\_mnist.md)
-    * [Mahalanobis outlier detection on KDD Cup ‘99 dataset](examples/od\_mahalanobis\_kddcup.md)
-    * [Time-series outlier detection using Prophet on weather data](examples/od\_prophet\_weather.md)
-    * [Seq2Seq time series outlier detection on ECG data](examples/od\_seq2seq\_ecg.md)
-    * [Time series outlier detection with Seq2Seq models on synthetic data](examples/od\_seq2seq\_synth.md)
-    * [Time series outlier detection with Spectral Residuals on synthetic data](examples/od\_sr\_synth.md)
-    * [VAE outlier detection for income prediction](examples/od\_vae\_adult.md)
-    * [VAE outlier detection on CIFAR10](examples/od\_vae\_cifar10.md)
-    * [VAE outlier detection on KDD Cup ‘99 dataset](examples/od\_vae\_kddcup.md)
+    * [AE outlier detection on CIFAR10](examples/od_ae_cifar10.md)
+    * [AEGMM and VAEGMM outlier detection on KDD Cup ‘99 dataset](examples/od_aegmm_kddcup.md)
+    * [Isolation Forest outlier detection on KDD Cup ‘99 dataset](examples/od_if_kddcup.md)
+    * [Likelihood Ratio Outlier Detection on Genomic Sequences](examples/od_llr_genome.md)
+    * [Likelihood Ratio Outlier Detection with PixelCNN++](examples/od_llr_mnist.md)
+    * [Mahalanobis outlier detection on KDD Cup ‘99 dataset](examples/od_mahalanobis_kddcup.md)
+    * [Time-series outlier detection using Prophet on weather data](examples/od_prophet_weather.md)
+    * [Seq2Seq time series outlier detection on ECG data](examples/od_seq2seq_ecg.md)
+    * [Time series outlier detection with Seq2Seq models on synthetic data](examples/od_seq2seq_synth.md)
+    * [Time series outlier detection with Spectral Residuals on synthetic data](examples/od_sr_synth.md)
+    * [VAE outlier detection for income prediction](examples/od_vae_adult.md)
+    * [VAE outlier detection on CIFAR10](examples/od_vae_cifar10.md)
+    * [VAE outlier detection on KDD Cup ‘99 dataset](examples/od_vae_kddcup.md)
 * [Drift Detection](cd/README.md)
   * [Methods](cd/methods/README.md)
     * [Offline](cd/methods/offline/README.md)
@@ -53,30 +52,30 @@
       * [Online Cramér-von Mises](cd/methods/onlinecvmdrift.md)
       * [Online Fisher’s Exact Test](cd/methods/onlinefetdrift.md)
   * [Examples](cd/examples/README.md)
-    * [Categorical and mixed type data drift detection on income prediction](examples/cd\_chi2ks\_adult.md)
-    * [Learned drift detectors on Adult Census](examples/cd\_clf\_adult.md)
-    * [Learned drift detectors on CIFAR-10](examples/cd\_clf\_cifar10.md)
-    * [Context-aware drift detection on news articles](examples/cd\_context\_20newsgroup.md)
-    * [Context-aware drift detection on ECGs](examples/cd\_context\_ecg.md)
-    * [Model Distillation drift detector on CIFAR-10](examples/cd\_distillation\_cifar10.md)
-    * [Kolmogorov-Smirnov data drift detector on CIFAR-10](examples/cd\_ks\_cifar10.md)
-    * [Maximum Mean Discrepancy drift detector on CIFAR-10](examples/cd\_mmd\_cifar10.md)
-    * [Scaling up drift detection with KeOps](examples/cd\_mmd\_keops.md)
-    * [Model uncertainty based drift detection on CIFAR-10 and Wine-Quality datasets](examples/cd\_model\_unc\_cifar10\_wine.md)
-    * [Drift detection on molecular graphs](examples/cd\_mol.md)
-    * [Online drift detection for Camelyon17 medical imaging dataset](examples/cd\_online\_camelyon.md)
-    * [Online Drift Detection on the Wine Quality Dataset](examples/cd\_online\_wine.md)
-    * [Interpretable drift detection with the spot-the-diff detector on MNIST and Wine-Quality datasets](examples/cd\_spot\_the\_diff\_mnist\_wine.md)
-    * [Supervised drift detection on the penguins dataset](examples/cd\_supervised\_penguins.md)
-    * [Drift detection on Amazon reviews](examples/cd\_text\_amazon.md)
-    * [Text drift detection on IMDB movie reviews](examples/cd\_text\_imdb.md)
+    * [Categorical and mixed type data drift detection on income prediction](examples/cd_chi2ks_adult.md)
+    * [Learned drift detectors on Adult Census](examples/cd_clf_adult.md)
+    * [Learned drift detectors on CIFAR-10](examples/cd_clf_cifar10.md)
+    * [Context-aware drift detection on news articles](examples/cd_context_20newsgroup.md)
+    * [Context-aware drift detection on ECGs](examples/cd_context_ecg.md)
+    * [Model Distillation drift detector on CIFAR-10](examples/cd_distillation_cifar10.md)
+    * [Kolmogorov-Smirnov data drift detector on CIFAR-10](examples/cd_ks_cifar10.md)
+    * [Maximum Mean Discrepancy drift detector on CIFAR-10](examples/cd_mmd_cifar10.md)
+    * [Scaling up drift detection with KeOps](examples/cd_mmd_keops.md)
+    * [Model uncertainty based drift detection on CIFAR-10 and Wine-Quality datasets](examples/cd_model_unc_cifar10_wine.md)
+    * [Drift detection on molecular graphs](examples/cd_mol.md)
+    * [Online drift detection for Camelyon17 medical imaging dataset](examples/cd_online_camelyon.md)
+    * [Online Drift Detection on the Wine Quality Dataset](examples/cd_online_wine.md)
+    * [Interpretable drift detection with the spot-the-diff detector on MNIST and Wine-Quality datasets](examples/cd_spot_the_diff_mnist_wine.md)
+    * [Supervised drift detection on the penguins dataset](examples/cd_supervised_penguins.md)
+    * [Drift detection on Amazon reviews](examples/cd_text_amazon.md)
+    * [Text drift detection on IMDB movie reviews](examples/cd_text_imdb.md)
 * [Adversarial Detection](adversarial-detection/README.md)
   * [Methods](adversarial-detection/methods/README.md)
     * [Adversarial Auto-Encoder](ad/methods/adversarialae.md)
     * [Model Distillation](ad/methods/modeldistillation.md)
   * [Examples](adversarial-detection/examples/README.md)
-    * [Adversarial AE detection and correction on CIFAR-10](examples/ad\_ae\_cifar10.md)
-* [Deployment](examples/alibi\_detect\_deploy.md)
+    * [Adversarial AE detection and correction on CIFAR-10](examples/ad_ae_cifar10.md)
+* [Deployment](examples/alibi_detect_deploy.md)
 * [Datasets](datasets/overview.md)
 * [Models](models/overview.md)
 * [Bibliography](bibliography.md)

docs-gb/ad/methods/modeldistillation.md

@@ -6,38 +6,35 @@ jupyter:
     name: python3
 ---
 
+# Model Distillation
+
 [source](../../api/alibi_detect.ad.model_distillation.rst)
 
-# Model distillation
+## Model distillation
 
-## Overview
+### Overview
 
-[Model distillation](https://arxiv.org/abs/1503.02531) is a technique that is used to transfer knowledge from a large network to a smaller network. Typically, it consists of training a second model with a simplified architecture on soft targets (the output distributions or the logits) obtained from the original model. 
+[Model distillation](https://arxiv.org/abs/1503.02531) is a technique that is used to transfer knowledge from a large network to a smaller network. Typically, it consists of training a second model with a simplified architecture on soft targets (the output distributions or the logits) obtained from the original model.
 
-Here, we apply model distillation to obtain harmfulness scores, by comparing the output distributions of the original model with the output distributions 
-of the distilled model, in order to detect adversarial data, malicious data drift or data corruption.
-We use the following definition of harmful and harmless data points:
+Here, we apply model distillation to obtain harmfulness scores, by comparing the output distributions of the original model with the output distributions of the distilled model, in order to detect adversarial data, malicious data drift or data corruption. We use the following definition of harmful and harmless data points:
 
 * Harmful data points are defined as inputs for which the model's predictions on the uncorrupted data are correct while the model's predictions on the corrupted data are wrong.
-
 * Harmless data points are defined as inputs for which the model's predictions on the uncorrupted data are correct and the model's predictions on the corrupted data remain correct.
 
-Analogously to the [adversarial AE detector](https://arxiv.org/abs/2002.09364), which is also part of the library, the model distillation detector picks up drift that reduces the performance of the classification model. 
+Analogously to the [adversarial AE detector](https://arxiv.org/abs/2002.09364), which is also part of the library, the model distillation detector picks up drift that reduces the performance of the classification model.
 
 The detector can be used as follows:
 
 * Given an input $x,$ an adversarial score $S(x)$ is computed. $S(x)$ equals the value loss function employed for distillation calculated between the original model's output and the distilled model's output on $x$.
-
 * If $S(x)$ is above a threshold (explicitly defined or inferred from training data), the instance is flagged as adversarial.
 
-## Usage
+### Usage
 
-### Initialize
+#### Initialize
 
 Parameters:
 
 * `threshold`: threshold value above which the instance is flagged as an adversarial instance.
-
 * `distilled_model`: `tf.keras.Sequential` instance containing the model used for distillation. Example:
 
 ```python
@@ -59,10 +56,8 @@ model = tf.keras.Model(inputs=inputs, outputs=outputs)
 ```
 
 * `loss_type`: type of loss used for distillation. Supported losses: 'kld', 'xent'.
-
 * `temperature`: Temperature used for model prediction scaling. Temperature <1 sharpens the prediction probability distribution which can be beneficial for prediction distributions with high entropy.
-
-* `data_type`: can specify data type added to metadata. E.g. *'tabular'* or *'image'*.
+* `data_type`: can specify data type added to metadata. E.g. _'tabular'_ or _'image'_.
 
 Initialized detector example:
 
@@ -76,55 +71,44 @@ ad = ModelDistillation(
 )
 ```
 
-### Fit
+#### Fit
 
 We then need to train the detector. The following parameters can be specified:
 
 * `X`: training batch as a numpy array.
-
 * `loss_fn`: loss function used for training. Defaults to the custom model distillation loss.
-
 * `optimizer`: optimizer used for training. Defaults to [Adam](https://arxiv.org/abs/1412.6980) with learning rate 1e-3.
-
 * `epochs`: number of training epochs.
-
 * `batch_size`: batch size used during training.
-
 * `verbose`: boolean whether to print training progress.
-
 * `log_metric`: additional metrics whose progress will be displayed if verbose equals True.
-
 * `preprocess_fn`: optional data preprocessing function applied per batch during training.
 
-
 ```python
 ad.fit(X_train, epochs=50)
 ```
 
-The threshold for the adversarial / harmfulness score can be set via ```infer_threshold```. We need to pass a batch of instances $X$ and specify what percentage of those we consider to be normal via `threshold_perc`. Even if we only have normal instances in the batch, it might be best to set the threshold value a bit lower (e.g. $95$%) since  the model could have misclassified training instances.
+The threshold for the adversarial / harmfulness score can be set via `infer_threshold`. We need to pass a batch of instances $X$ and specify what percentage of those we consider to be normal via `threshold_perc`. Even if we only have normal instances in the batch, it might be best to set the threshold value a bit lower (e.g. $95$%) since the model could have misclassified training instances.
 
 ```python
 ad.infer_threshold(X_train, threshold_perc=95, batch_size=64)
 ```
 
-### Detect
+#### Detect
 
 We detect adversarial / harmful instances by simply calling `predict` on a batch of instances `X`. We can also return the instance level score by setting `return_instance_score` to True.
 
 The prediction takes the form of a dictionary with `meta` and `data` keys. `meta` contains the detector's metadata while `data` is also a dictionary which contains the actual predictions stored in the following keys:
 
-* `is_adversarial`: boolean whether instances are above the threshold and therefore adversarial instances. The array is of shape *(batch size,)*.
-
+* `is_adversarial`: boolean whether instances are above the threshold and therefore adversarial instances. The array is of shape _(batch size,)_.
 * `instance_score`: contains instance level scores if `return_instance_score` equals True.
 
-
 ```python
 preds_detect = ad.predict(X, batch_size=64, return_instance_score=True)
 ```
 
-## Examples
+### Examples
 
-### Image
+#### Image
 
 [Harmful drift detection through model distillation on CIFAR10](../../examples/cd_distillation_cifar10.ipynb)
-