Fix documentation for new syntax on GAN synthesizers (#479)

* Syntax for GAN calls

Co-authored-by: Joshua <[email protected]>

synth/README.md

@@ -6,6 +6,7 @@
 
 Differentially private synthesizers for tabular data.  Package includes:
 * MWEM
+* MST
 * QUAIL
 * DP-CTGAN
 * PATE-CTGAN
@@ -19,66 +20,11 @@ pip install smartnoise-synth
 
 ## Using
 
-### MWEM
-
-```python
-import snsynth
-import pandas as pd
-import numpy as np
-
-pums = pd.read_csv(pums_csv_path, index_col=None) # in datasets/
-pums = pums.drop(['income'], axis=1)
-nf = pums.to_numpy().astype(int)
-
-synth = snsynth.MWEMSynthesizer(epsilon=1.0, split_factor=nf.shape[1]) 
-synth.fit(nf)
-
-sample = synth.sample(10)
-print(sample)
-```
-### DP-CTGAN
-
-```python
-import snsynth
-import pandas as pd
-import numpy as np
-
-from snsynth.pytorch.nn import DPCTGAN
-from snsynth.pytorch import PytorchDPSynthesizer
-
-pums = pd.read_csv(pums_csv_path, index_col=None) # in datasets/
-pums = pums.drop(['income'], axis=1)
-
-synth = PytorchDPSynthesizer(1.0, DPCTGAN(), None)
-synth.fit(pums, categorical_columns=pums.columns)
-
-sample = synth.sample(10) # synthesize 10 rows
-print(sample)
-```
-
-### PATE-CTGAN
-
-```python
-import snsynth
-import pandas as pd
-import numpy as np
-
-from snsynth.pytorch.nn import PATECTGAN
-from snsynth.pytorch import PytorchDPSynthesizer
-
-pums = pd.read_csv(pums_csv_path, index_col=None) # in datasets/
-pums = pums.drop(['income'], axis=1)
-
-synth = PytorchDPSynthesizer(1.0, PATECTGAN(regularization='dragan'), None)
-synth.fit(pums, categorical_columns=pums.columns)
-
-sample = synth.sample(10) # synthesize 10 rows
-print(sample)
-```
+Please see the [SmartNoise synthesizers documentation](https://docs.smartnoise.org/synth/index.html) for usage examples.
 
 ## Note on Inputs
 
-MWEM, DP-CTGAN, and PATE-CTGAN require columns to be categorical. If you have columns with continuous values, you should discretize them before fitting.  Take care to discretize in a way that does not reveal information about the distribution of the data.
+MWEM and MST require columns to be categorical. If you have columns with continuous values, you should discretize them before fitting.  Take care to discretize in a way that does not reveal information about the distribution of the data.
 
 ## Communication
 

synth/docs/source/index.rst

@@ -20,7 +20,7 @@ Getting Started
 MWEM
 ----
 
-Multiplicative Weights Exponential Mechanism.
+Multiplicative Weights Exponential Mechanism.  From "`A Simple and Practical Algorithm for Differentially Private Data Release <https://www.cs.huji.ac.il/~katrina//papers/mwem-nips.pdf>`_".
 
 .. code-block:: python
 
@@ -41,49 +41,76 @@ Multiplicative Weights Exponential Mechanism.
 DP-CTGAN
 --------
 
-Conditional tabular GAN with differentially private stochastic gradient descent.
+Conditional tabular GAN with differentially private stochastic gradient descent.  From "`Modeling Tabular data using Conditional GAN <https://arxiv.org/abs/1907.00503>`_".
 
 .. code-block:: python
 
-  imprt snsynth
+  import snsynth
   import pandas as pd
   import numpy as np
 
   from snsynth.pytorch.nn import DPCTGAN
   from snsynth.pytorch import PytorchDPSynthesizer
+  from snsynth.preprocessors.data_transformer import BaseTransformer
 
   pums = pd.read_csv(pums_csv_path, index_col=None) # in datasets/
   pums = pums.drop(['income'], axis=1)
 
-  synth = PytorchDPSynthesizer(1.0, DPCTGAN(), None)
-  synth.fit(pums, categorical_columns=pums.columns)
+  synth = PytorchDPSynthesizer(1.0, DPCTGAN())
+  synth.fit(pums, categorical_columns=list(pums.columns), transformer=BaseTransformer)
 
   sample = synth.sample(10) # synthesize 10 rows
   print(sample)
 
 PATE-CTGAN
 ----------
 
-Conditional tabular GAN using Private Aggregation of Teacher Ensembles.
+Conditional tabular GAN using Private Aggregation of Teacher Ensembles.  From "`PATE-GAN: Generating Synthetic Data with Differential Privacy Guarantees <https://openreview.net/pdf?id=S1zk9iRqF7>`_" and "`Modeling Tabular data using Conditional GAN <https://arxiv.org/abs/1907.00503>`_".
 
 .. code-block:: python
 
-  imprt snsynth
+  import snsynth
   import pandas as pd
   import numpy as np
 
   from snsynth.pytorch.nn import PATECTGAN
   from snsynth.pytorch import PytorchDPSynthesizer
+  from snsynth.preprocessors.data_transformer import BaseTransformer
+
+  pums = pd.read_csv(pums_csv_path, index_col=None) # in datasets/
+  pums = pums.drop(['income'], axis=1)
+
+  synth = PytorchDPSynthesizer(1.0, PATECTGAN(regularization='dragan'))
+  synth.fit(pums, categorical_columns=list(pums.columns), transformer=BaseTransformer)
+
+  sample = synth.sample(10) # synthesize 10 rows
+  print(sample)
+
+
+MST
+---
+
+MST achieves state of the art results for marginals over categorical data, and does well even with small source data.  From McKenna et al. "`Winning the NIST Contest: A scalable and general approach to differentially private synthetic data <https://arxiv.org/abs/2108.04978>`_"
+
+.. code-block:: python
+
+  import snsynth
+  import pandas as pd
+  import numpy as np
 
   pums = pd.read_csv(pums_csv_path, index_col=None) # in datasets/
   pums = pums.drop(['income'], axis=1)
 
-  synth = PytorchDPSynthesizer(1.0, PATECTGAN(regularization='dragan'), None)
-  synth.fit(pums, categorical_columns=pums.columns)
+  Domains = {"pums": "samples/mst_sample/pums-domain.json"} # in samples/mst_sample
+  synth = snsynth.MSTSynthesizer(domains_dict=Domains, domain="pums", epsilon=1.0)
+  synth.fit(pums)
 
   sample = synth.sample(10) # synthesize 10 rows
   print(sample)
 
+For more, see the `sample notebook <https://github.com/opendp/smartnoise-sdk/tree/main/synth/samples/mst_sample>`_
+
+
 This is version |version| of the guides, last built on |today|.
 
 .. |opendp-logo| image:: _static/images/opendp-logo.png