Anchor text (#15)

* anchor text

* clean up notebook

* anchor text

* clean up notebook

* add anchor text

* update example notebook

* add build docs

* output example notebook

* update example

* change example title

* add logging and adjust UNK data type

alibi/datasets.py

@@ -1,7 +1,36 @@
+from io import BytesIO
 import numpy as np
 import pandas as pd
 from sklearn.preprocessing import LabelEncoder
+import tarfile
 from typing import Tuple
+from urllib.request import urlopen
+
+
+def movie_sentiment() -> Tuple[list, list]:
+    """
+    The movie review dataset, equally split between negative and positive reviews.
+
+    Returns
+    -------
+    Movie reviews and sentiment labels (0 means 'negative' and 1 means 'positive').
+    """
+    url = 'http://www.cs.cornell.edu/People/pabo/movie-review-data/rt-polaritydata.tar.gz'
+    resp = urlopen(url)
+    tar = tarfile.open(fileobj=BytesIO(resp.read()), mode="r:gz")
+    data = []
+    labels = []
+    for i, member in enumerate(tar.getnames()[1:]):
+        f = tar.extractfile(member)
+        for line in f.readlines():
+            try:
+                line.decode('utf8')
+            except UnicodeDecodeError:
+                continue
+            data.append(line.decode('utf8').strip())
+            labels.append(i)
+    tar.close()
+    return data, labels
 
 
 def adult(features_drop: list = ["fnlwgt", "Education-Num"]) -> Tuple[np.ndarray, np.ndarray, list, dict]:

alibi/explainers/__init__.py

@@ -3,5 +3,7 @@
 """
 
 from .anchor.anchor_tabular import AnchorTabular
+from .anchor.anchor_text import AnchorText
 
-__all__ = ["AnchorTabular"]
+__all__ = ["AnchorTabular",
+           "AnchorText"]

alibi/explainers/anchor/anchor_base.py

@@ -296,7 +296,7 @@ def make_tuples(previous_best: list, state: dict) -> list:
         return list(new_tuples)
 
     @staticmethod
-    def get_sample_fns(sample_fn: Callable, tuples: list, state: dict) -> list:
+    def get_sample_fns(sample_fn: Callable, tuples: list, state: dict, data_type: str = None) -> list:
         """
         Parameters
         ----------
@@ -306,12 +306,13 @@ def get_sample_fns(sample_fn: Callable, tuples: list, state: dict) -> list:
             List of anchor candidates
         state
             Dictionary with the relevant metrics like coverage and samples for candidate anchors
+        data_type
+            Data type for raw data
 
         Returns
         -------
         List with sample functions for each candidate anchor.
         """
-
         def complete_sample_fn(t: tuple, n: int) -> int:
             """
             Parameters
@@ -339,8 +340,9 @@ def complete_sample_fn(t: tuple, n: int) -> int:
                 prealloc_size = state['prealloc_size']
                 current_idx = data.shape[0]
                 state['data'] = np.vstack((state['data'], np.zeros((prealloc_size, data.shape[1]), data.dtype)))
+                dtype = data_type if data_type is not None else raw_data.dtype
                 state['raw_data'] = np.vstack((state['raw_data'], np.zeros((prealloc_size, raw_data.shape[1]),
-                                                                           raw_data.dtype)))
+                                                                           dtype=dtype)))
                 state['labels'] = np.hstack((state['labels'], np.zeros(prealloc_size, labels.dtype)))
             return labels.sum()
 
@@ -417,10 +419,10 @@ def get_anchor_from_tuple(t: tuple, state: dict) -> dict:
 
     @staticmethod
     def anchor_beam(sample_fn: Callable, delta: float = 0.05, epsilon: float = 0.1, batch_size: int = 10,
-                    desired_confidence: float = 1, beam_size: int = 1,
-                    verbose: bool = False, epsilon_stop: float = 0.05, min_samples_start: int = 0,
-                    max_anchor_size: int = None, verbose_every: int = 1,
-                    stop_on_first: bool = False, coverage_samples: int = 10000) -> dict:
+                    desired_confidence: float = 1, beam_size: int = 1, verbose: bool = False,
+                    epsilon_stop: float = 0.05, min_samples_start: int = 0, max_anchor_size: int = None,
+                    verbose_every: int = 1, stop_on_first: bool = False, coverage_samples: int = 10000,
+                    data_type: str = None) -> dict:
         """
         Parameters
         ----------
@@ -446,6 +448,8 @@ def anchor_beam(sample_fn: Callable, delta: float = 0.05, epsilon: float = 0.1,
             Whether to print intermediate output every verbose_every steps
         stop_on_first
         coverage_samples
+        data_type
+            Data type for raw data
 
         Returns
         -------
@@ -489,7 +493,8 @@ def anchor_beam(sample_fn: Callable, delta: float = 0.05, epsilon: float = 0.1,
         prealloc_size = batch_size * 10000
         current_idx = data.shape[0]
         data = np.vstack((data, np.zeros((prealloc_size, data.shape[1]), data.dtype)))
-        raw_data = np.vstack((raw_data, np.zeros((prealloc_size, raw_data.shape[1]), raw_data.dtype)))
+        dtype = data_type if data_type is not None else raw_data.dtype
+        raw_data = np.vstack((raw_data, np.zeros((prealloc_size, raw_data.shape[1]), dtype=dtype)))
         labels = np.hstack((labels, np.zeros(prealloc_size, labels.dtype)))
         n_features = data.shape[1]
         state = {'t_idx': collections.defaultdict(lambda: set()),
@@ -531,7 +536,7 @@ def anchor_beam(sample_fn: Callable, delta: float = 0.05, epsilon: float = 0.1,
             # these functions sample randomly for all features except for the ones in the candidate anchors
             # for the features in the anchor it uses the same category (categorical features) or samples from ...
             # ... the same bin (discretized numerical features) as the feature in the observation that is explained
-            sample_fns = AnchorBaseBeam.get_sample_fns(sample_fn, tuples, state)
+            sample_fns = AnchorBaseBeam.get_sample_fns(sample_fn, tuples, state, data_type=dtype)
 
             # for each tuple, get initial nb of samples used and prec(A)
             initial_stats = AnchorBaseBeam.get_initial_statistics(tuples, state)
@@ -606,7 +611,7 @@ def anchor_beam(sample_fn: Callable, delta: float = 0.05, epsilon: float = 0.1,
             tuples = []
             for i in range(0, current_size):
                 tuples.extend(best_of_size[i])
-            sample_fns = AnchorBaseBeam.get_sample_fns(sample_fn, tuples, state)
+            sample_fns = AnchorBaseBeam.get_sample_fns(sample_fn, tuples, state, data_type=dtype)
             initial_stats = AnchorBaseBeam.get_initial_statistics(tuples, state)
             chosen_tuples = AnchorBaseBeam.lucb(sample_fns, initial_stats, epsilon,
                                                 delta, batch_size, 1, verbose=verbose)