diff --git a/CHANGES.rst b/CHANGES.rst
index 20a98c85..92588f7a 100644
--- a/CHANGES.rst
+++ b/CHANGES.rst
@@ -1,6 +1,18 @@
 Changelog
 =========
 
+0.10.1 (2019-08-29)
+-------------------
+
+* Don't include typing dependency on Python 3.5+
+  to fix installation on Python 3.7
+
+0.10.0 (2019-08-21)
+-------------------
+
+* Keras image classifiers: explaining predictions with Grad-CAM
+  (GSoC-2019 project by @teabolt).
+
 0.9.0 (2019-07-05)
 ------------------
 
diff --git a/eli5/__init__.py b/eli5/__init__.py
index 416681eb..1b125976 100644
--- a/eli5/__init__.py
+++ b/eli5/__init__.py
@@ -1,7 +1,7 @@
 # -*- coding: utf-8 -*-
 from __future__ import absolute_import
 
-__version__ = '0.9.0'
+__version__ = '0.10.1'
 
 from .formatters import (
     format_as_html,
@@ -96,6 +96,7 @@
 except ImportError:
     # keras is not available
     pass
+<<<<<<< HEAD
 
 
 try:
@@ -104,4 +105,6 @@
     )
 except ImportError:
     # pytorch is not available
-    pass
\ No newline at end of file
+    pass
+=======
+>>>>>>> ec0f51c60aaf360327ca18e3e0cdae2222cec6bf
diff --git a/eli5/_feature_names.py b/eli5/_feature_names.py
index fecb820f..ff1fd80c 100644
--- a/eli5/_feature_names.py
+++ b/eli5/_feature_names.py
@@ -5,8 +5,8 @@
     Union, Callable, Pattern
 )
 
-import numpy as np  # type: ignore
-import scipy.sparse as sp  # type: ignore
+import numpy as np
+import scipy.sparse as sp
 
 
 class FeatureNames(Sized, Iterable):
diff --git a/eli5/_feature_weights.py b/eli5/_feature_weights.py
index 0b737795..1c096e3a 100644
--- a/eli5/_feature_weights.py
+++ b/eli5/_feature_weights.py
@@ -1,7 +1,7 @@
 # -*- coding: utf-8 -*-
 from __future__ import absolute_import
 
-import numpy as np  # type: ignore
+import numpy as np
 
 from eli5.base import FeatureWeights, FeatureWeight
 from .utils import argsort_k_largest_positive, argsort_k_smallest, mask
diff --git a/eli5/_graphviz.py b/eli5/_graphviz.py
index 4632a925..36b26a17 100644
--- a/eli5/_graphviz.py
+++ b/eli5/_graphviz.py
@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-import graphviz  # type: ignore
+import graphviz
 
 
 def is_supported():
diff --git a/eli5/base.py b/eli5/base.py
index 1f120fef..ef50fb0a 100644
--- a/eli5/base.py
+++ b/eli5/base.py
@@ -1,7 +1,7 @@
 # -*- coding: utf-8 -*-
 from typing import Any, List, Tuple, Union, Optional
 
-import numpy as np # type: ignore
+import numpy as np
 
 from .base_utils import attrs
 from .formatters.features import FormattedFeatureName
diff --git a/eli5/base_utils.py b/eli5/base_utils.py
index 779c6d64..f1081c3b 100644
--- a/eli5/base_utils.py
+++ b/eli5/base_utils.py
@@ -1,9 +1,9 @@
 import inspect
 
-import attr  # type: ignore
+import attr
 
 try:
-    from functools import singledispatch  # type: ignore
+    from functools import singledispatch
 except ImportError:
     from singledispatch import singledispatch  # type: ignore
 
diff --git a/eli5/catboost.py b/eli5/catboost.py
index 2e495cb6..56abfa9e 100644
--- a/eli5/catboost.py
+++ b/eli5/catboost.py
@@ -1,7 +1,7 @@
 from __future__ import absolute_import, division
 
-import numpy as np  # type: ignore
-import catboost # type: ignore
+import numpy as np
+import catboost
 
 from eli5.explain import explain_weights
 from eli5._feature_importances import get_feature_importance_explanation
diff --git a/eli5/formatters/as_dataframe.py b/eli5/formatters/as_dataframe.py
index 6f2fc302..5b801e75 100644
--- a/eli5/formatters/as_dataframe.py
+++ b/eli5/formatters/as_dataframe.py
@@ -2,7 +2,7 @@
 from typing import Any, Dict, List, Optional
 import warnings
 
-import pandas as pd  # type: ignore
+import pandas as pd
 
 import eli5
 from eli5.base import (
diff --git a/eli5/formatters/as_dict.py b/eli5/formatters/as_dict.py
index 1878d8b0..fbad5ee5 100644
--- a/eli5/formatters/as_dict.py
+++ b/eli5/formatters/as_dict.py
@@ -1,7 +1,7 @@
 import six
 
-import attr   # type: ignore
-import numpy as np   # type: ignore
+import attr
+import numpy as np
 
 from .features import FormattedFeatureName
 
diff --git a/eli5/formatters/html.py b/eli5/formatters/html.py
index 54cf2e6c..0167b33e 100644
--- a/eli5/formatters/html.py
+++ b/eli5/formatters/html.py
@@ -3,8 +3,8 @@
 from itertools import groupby
 from typing import List, Optional, Tuple
 
-import numpy as np  # type: ignore
-from jinja2 import Environment, PackageLoader  # type: ignore
+import numpy as np
+from jinja2 import Environment, PackageLoader
 
 from eli5 import _graphviz
 from eli5.base import (Explanation, TargetExplanation, FeatureWeights,
diff --git a/eli5/formatters/image.py b/eli5/formatters/image.py
index e2f15684..5175961f 100644
--- a/eli5/formatters/image.py
+++ b/eli5/formatters/image.py
@@ -2,11 +2,14 @@
 from __future__ import absolute_import
 from typing import Union, Optional, Callable
 
-import numpy as np # type: ignore
-from PIL import Image # type: ignore
-import matplotlib.cm # type: ignore
+import numpy as np
+from PIL import Image
+import matplotlib.cm
 
 from eli5.base import Explanation
+from eli5.nn.gradcam import (
+    _validate_heatmap,
+)
 
 
 def format_as_image(expl, # type: Explanation
@@ -287,14 +290,6 @@ def _validate_image(image):
                         'Got: {}'.format(image))
 
 
-def _validate_heatmap(heatmap):
-    # type: (np.ndarray) -> None
-    """Check that ``heatmap`` has the right type."""
-    if not isinstance(heatmap, np.ndarray):
-        raise TypeError('heatmap must be a numpy.ndarray instance. '
-                        'Got: {}'.format(heatmap))
-
-
 def _needs_normalization(heatmap):
     # type: (np.ndarray) -> bool
     """Return whether ``heatmap`` values are in the interval [0, 1]."""
@@ -311,4 +306,4 @@ def _normalize_heatmap(h, epsilon=1e-07):
     # https://datascience.stackexchange.com/questions/5885/how-to-scale-an-array-of-signed-integers-to-range-from-0-to-1
     # add eps to avoid division by zero in case heatmap is all 0's
     # this also means that lmap max will be slightly less than the 'true' max
-    return (h - h.min()) / (h.max() - h.min() + epsilon)
+    return (h - h.min()) / (h.max() - h.min() + epsilon)
\ No newline at end of file
diff --git a/eli5/formatters/text.py b/eli5/formatters/text.py
index a6269ed3..e6abb286 100644
--- a/eli5/formatters/text.py
+++ b/eli5/formatters/text.py
@@ -2,6 +2,7 @@
 from __future__ import absolute_import
 from itertools import chain
 import six
+from tabulate import tabulate
 from typing import List, Optional, Iterator
 
 from eli5.base import Explanation, FeatureImportances
@@ -9,7 +10,8 @@
 from .features import FormattedFeatureName
 from .utils import (
     format_signed, format_value, format_weight, has_any_values_for_weights,
-    replace_spaces, should_highlight_spaces, tabulate)
+    replace_spaces, should_highlight_spaces)
+from .utils import tabulate as eli5_tabulate
 from .trees import tree2text
 
 
@@ -153,7 +155,6 @@ def _decision_tree_lines(explanation):
 
 def _transition_features_lines(explanation):
     # type: (Explanation) -> List[str]
-    from tabulate import tabulate  # type: ignore
     tf = explanation.transition_features
     assert tf is not None
     return [
@@ -203,7 +204,7 @@ def _targets_lines(explanation,  # type: Explanation
 
         w = target.feature_weights
         assert w is not None
-        table = tabulate(
+        table = eli5_tabulate(
             [table_line(fw) for fw in chain(w.pos, reversed(w.neg))],
             header=table_header,
             col_align=col_align,
diff --git a/eli5/formatters/text_helpers.py b/eli5/formatters/text_helpers.py
index dc5ff28a..c63c66b2 100644
--- a/eli5/formatters/text_helpers.py
+++ b/eli5/formatters/text_helpers.py
@@ -1,7 +1,7 @@
 from collections import Counter
 from typing import List, Optional
 
-import numpy as np  # type: ignore
+import numpy as np
 
 from eli5.base import TargetExplanation, WeightedSpans, DocWeightedSpans
 from eli5.base_utils import attrs
diff --git a/eli5/formatters/trees.py b/eli5/formatters/trees.py
index 7be1e9c3..5cfbee10 100644
--- a/eli5/formatters/trees.py
+++ b/eli5/formatters/trees.py
@@ -24,6 +24,8 @@ def p(*args):
         else:
             assert node.left is not None
             assert node.right is not None
+            assert node.threshold is not None
+
             feat_name = node.feature_name
 
             if depth > 0:
diff --git a/eli5/formatters/utils.py b/eli5/formatters/utils.py
index 542402d1..2e6d2d39 100644
--- a/eli5/formatters/utils.py
+++ b/eli5/formatters/utils.py
@@ -2,10 +2,9 @@
 from itertools import chain
 import re
 import six
-from numbers import Real
 from typing import Any, Union, List, Dict, Callable, Match, Optional
 
-import numpy as np  # type: ignore
+import numpy as np
 
 from eli5.base import Explanation
 from .features import FormattedFeatureName
@@ -143,12 +142,12 @@ def tabulate(data,  # type: List[List[Any]]
 
 
 def format_weight(value):
-    # type: (Real) -> str
+    # type: (float) -> str
     return '{:+.3f}'.format(value)
 
 
 def format_value(value):
-    # type: (Optional[Real]) -> str
+    # type: (Optional[float]) -> str
     if value is None:
         return ''
     elif np.isnan(value):
diff --git a/eli5/ipython.py b/eli5/ipython.py
index 0039c39e..033a840c 100644
--- a/eli5/ipython.py
+++ b/eli5/ipython.py
@@ -3,7 +3,7 @@
 from typing import Any, Dict, Tuple
 import warnings
 
-from IPython.display import HTML, Image  # type: ignore
+from IPython.display import HTML, Image
 
 from .explain import explain_weights, explain_prediction
 from .formatters import format_as_html, fields
@@ -11,7 +11,7 @@
     from .formatters.image import format_as_image
 except ImportError as e:
     # missing dependencies
-    format_as_image = e # type: ignore
+    format_as_image = e  # type: ignore
 
 
 FORMAT_KWARGS = {'include_styles', 'force_weights',
diff --git a/eli5/keras/explain_prediction.py b/eli5/keras/explain_prediction.py
index 6d465f34..1613af72 100644
--- a/eli5/keras/explain_prediction.py
+++ b/eli5/keras/explain_prediction.py
@@ -2,14 +2,14 @@
 from __future__ import absolute_import
 from typing import Union, Optional, Callable, List, Tuple, Generator, TYPE_CHECKING
 if TYPE_CHECKING:
-    import PIL # type: ignore
-
-import numpy as np # type: ignore
-import keras # type: ignore
-import keras.backend as K # type: ignore
-from keras.models import Model # type: ignore
-from keras.layers import Layer # type: ignore
-from keras.layers import (  # type: ignore
+    import PIL
+
+import numpy as np
+import keras
+import keras.backend as K
+from keras.models import Model
+from keras.layers import Layer
+from keras.layers import (
     Conv2D,
     MaxPooling2D,
     AveragePooling2D,
@@ -26,7 +26,7 @@
     GRU,
     Bidirectional,
 )
-from keras.preprocessing.image import array_to_img # type: ignore
+from keras.preprocessing.image import array_to_img
 
 from eli5.base import (
     Explanation, 
@@ -36,12 +36,9 @@
 from eli5.nn.gradcam import (
     gradcam_heatmap,
     DESCRIPTION_GRADCAM,
-    _validate_targets,
-    _validate_classification_target,
 )
 from eli5.nn.text import (
-    gradcam_text_spans,
-    _is_character_tokenization,
+    gradcam_spans,
 )
 from .gradcam import (
     gradcam_backend_keras,
@@ -248,7 +245,7 @@ def explain_prediction_keras_image(model,
           * ``target`` ID of target class.
           * ``score`` value for predicted class.
     """
-    _validate_params(model, doc, targets=targets)
+    _validate_params(model, doc)
     if image is None:
         image = _extract_image(doc)
 
@@ -354,7 +351,7 @@ def explain_prediction_keras_text(model,
 
     """
     assert tokens is not None
-    _validate_params(model, doc, targets=targets, tokens=tokens)
+    _validate_params(model, doc)
     tokens = _unbatch_tokens(tokens)
 
     if layer is not None:
@@ -373,13 +370,13 @@ def explain_prediction_keras_text(model,
     predicted_idx, = predicted_idx
     predicted_val, = predicted_val
     heatmap, = heatmap
-    text_vals = gradcam_text_spans(heatmap,
-                                   tokens,
-                                   doc,
-                                   pad_value=pad_value,
-                                   pad_token=pad_token,
-                                   interpolation_kind=interpolation_kind,
-                                   )
+    text_vals = gradcam_spans(heatmap,
+                              tokens,
+                              doc,
+                              pad_value=pad_value,
+                              pad_token=pad_token,
+                              interpolation_kind=interpolation_kind,
+                              )
     # TODO: padding could be relevant for images too?
     tokens, heatmap, weighted_spans = text_vals
     return Explanation(
@@ -553,18 +550,11 @@ def _backward_layers(model):
 
 def _validate_params(model, # type: Model
                      doc, # type: np.ndarray
-                     targets=None, # type: Optional[list]
-                     tokens=None, # type: Optional[Union[np.ndarray, list]]
                      ):
     # type: (...) -> None
-    """Helper for validating all explanation function parameters."""
+    """Helper for validating explanation function parameters."""
     _validate_model(model)
     _validate_doc(doc)
-    if targets is not None:
-        _validate_targets(targets)
-        _validate_classification_target(targets[0], model.output_shape)
-    if tokens is not None:
-        _validate_tokens(doc, tokens)
 
 
 def _validate_model(model):
@@ -589,59 +579,4 @@ def _validate_doc(doc):
         raise ValueError('"doc" batch size must be 1. '
                          'Got doc with batch size: %d' % batch_size)
 
-    # Note that validation of the input shape, etc is done by Keras
-
-
-# FIXME: break this function up
-def _validate_tokens(doc, tokens):
-    # type: (np.ndarray, Union[np.ndarray, list]) -> None
-    """Check that ``tokens`` contains correct items and matches ``doc``."""
-    batch_size, doc_len = doc.shape
-    if not isinstance(tokens, (list, np.ndarray)):
-        # wrong type
-        raise TypeError('"tokens" must be list or numpy.ndarray. '
-                        'Got "{}".'.format(tokens))
-
-    if len(tokens) == 0:
-        # empty list
-        raise ValueError('"tokens" is empty: {}'.format(tokens))
-
-    an_entry = tokens[0]
-    if isinstance(an_entry, str):
-        # no batch
-        if batch_size != 1:
-            # doc is batched but tokens is not
-            raise ValueError('If passing "tokens" without batch dimension, '
-                             '"doc" must have batch size = 1.'
-                             'Got "doc" with batch size = %d.' % batch_size)
-        tokens_len = len(tokens)
-    elif isinstance(an_entry, (list, np.ndarray)):
-        # batched
-        tokens_batch_size = len(tokens)
-        if tokens_batch_size != batch_size:
-            # batch lengths do not match
-            raise ValueError('"tokens" must have same number of samples '
-                             'as in doc batch. Got: "tokens" samples: %d, '
-                             'doc samples: %d' % (tokens_batch_size, batch_size))
-
-        a_token = an_entry[0]
-        if not isinstance(a_token, str):
-            # actual contents are not strings
-            raise TypeError('Second axis in "tokens" must contain strings. '
-                             'Found "{}" (type "{}")'.format(a_token, type(a_token)))
-
-        # https://stackoverflow.com/a/35791116/11555448
-        it = iter(tokens)
-        the_len = len(next(it))
-        if not all(len(l) == the_len for l in it):
-            raise ValueError('"tokens" samples do not have the same length.')
-        tokens_len = the_len
-    else:
-        raise TypeError('"tokens" must be an array of strings, '
-                        'or an array of string arrays. '
-                        'Got "{}".'.format(tokens))
-
-    if tokens_len != doc_len:
-        raise ValueError('"tokens" and "doc" lengths must match. '
-                         '"tokens" length: "%d". "doc" length: "%d"'
-                         % (tokens_len, doc_len))
+    # Note that validation of the input shape, etc is done by Keras
\ No newline at end of file
diff --git a/eli5/keras/gradcam.py b/eli5/keras/gradcam.py
index c5f06fd2..5fd09526 100644
--- a/eli5/keras/gradcam.py
+++ b/eli5/keras/gradcam.py
@@ -7,24 +7,29 @@
 """
 
 from __future__ import absolute_import
-from typing import Optional, Tuple, List
+from typing import Any, Optional, Tuple, List
 
-import numpy as np # type: ignore
-import keras # type: ignore
-import keras.backend as K # type: ignore
-from keras.models import Model # type: ignore
-from keras.layers import Layer # type: ignore
+import numpy as np
+import keras
+import keras.backend as K
+from keras.models import Model
+from keras.layers import Layer
+
+from eli5.nn.gradcam import (
+    _validate_targets,
+    _validate_classification_target,
+)
 
 
 def gradcam_backend_keras(model, # type: Model
-    doc, # type: np.ndarray
-    targets, # type: Optional[List[int]]
-    activation_layer, # type: Layer
-    ):
+                          doc, # type: np.ndarray
+                          targets, # type: Optional[List[int]]
+                          activation_layer, # type: Layer
+                          ):
     # type: (...) -> Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]
     """
     Compute the terms and by-products required by the Grad-CAM formula.
-    
+
     Parameters
     ----------
     model : keras.models.Model
@@ -51,24 +56,13 @@ def gradcam_backend_keras(model, # type: Model
         (numpy.ndarray, np.ndarray, np.ndarray, np.ndarray)
         Values of variables.
     """
-    # score for class in targets
-    # TODO: maybe do the sum / loss calculation in this function and pass it to gradcam.
-    # This would be consistent with what is done in
-    # https://github.com/ramprs/grad-cam/blob/master/misc/utils.lua
-    # and https://github.com/ramprs/grad-cam/blob/master/classification.lua
-    # TODO: as in pytorch PR, separate out classification tensor code
-    if targets is not None:
-        target, = targets
-        predicted_idx = K.constant([target], dtype='int64')
-    else:
-        predicted_idx = _autoget_target_prediction(model)
-    # access value by index
-    predicted_val = K.gather(model.output[0, :], predicted_idx)
+    # class score
+    predicted_idx, predicted_val = _classification_target(model, targets)
 
     # output of target activation layer, i.e. activation maps of a conv layer
     activation_output = activation_layer.output
 
-    # score for class w.r.p.t. activation layer
+    # score w.r.p.t. activation layer
     grads = _calc_gradient(predicted_val, [activation_output])
 
     # TODO: gradcam on input layer
@@ -85,12 +79,12 @@ def gradcam_backend_keras(model, # type: Model
 
 
 def _calc_gradient(ys, xs):
-    # (K.variable, list) -> K.variable
-    # FIXME: K.variable is not the right type to use?
+    # (Any, list) -> Any
+    # TODO: In the future we can replace the annotation Any with a tensor type in Keras backend
     """
     Return the gradient of scalar ``ys`` with respect to each of list ``xs``,
     (must be singleton)
-    and apply grad normalization.
+    and apply gradient normalization.
     """
     # differentiate ys (scalar) with respect to each variable in xs
     # K.gradients tends to produce bigger values than tf.gradients
@@ -116,9 +110,43 @@ def _calc_gradient(ys, xs):
     return grads
 
 
-def _autoget_target_prediction(model):
-    # type: (Model) -> K.variable
-    """Automatically get the index with 
-    the highest predicted output from ``model``"""
-    output = model.output
-    return K.argmax(output, axis=-1)
\ No newline at end of file
+def _classification_target(model, targets):
+    # type: (Model, Optional[List[int]]) -> Tuple[Any, Any]
+    """Get a predicted index and its value from a classification based model."""
+    # TODO: maybe pass the loss/score to the gradcam function.
+    # This would be consistent with what is done in
+    # https://github.com/ramprs/grad-cam/blob/master/misc/utils.lua
+    # and https://github.com/ramprs/grad-cam/blob/master/classification.lua
+    if targets is not None:
+        _validate_targets(targets)
+        target, = targets
+        _validate_classification_target(target, model.output_shape)
+        # make a dummy index
+        predicted_idx = K.constant([target], dtype='int64')
+    else:
+        # take the index with the highest value
+        # from the array of predictions
+        predicted_idx = K.argmax(model.output, axis=-1)
+
+<<<<<<< HEAD
+    # access value by index
+    predicted_val = K.gather(model.output[0, :], predicted_idx)
+    return predicted_idx, predicted_val
+=======
+def _validate_target(target, output_shape):
+    # type: (int, tuple) -> None
+    """
+    Check whether ``target``, 
+    an integer index into the model's output
+    is valid for the given ``output_shape``.
+    """
+    if isinstance(target, int):
+        output_nodes = output_shape[1:][0]
+        if not (0 <= target < output_nodes):
+            raise ValueError('Prediction target index is ' 
+                             'outside the required range [0, {}). '
+                             'Got {}'.format(output_nodes, target))
+    else:
+        raise TypeError('Prediction target must be int. '
+                        'Got: {}'.format(target))
+>>>>>>> 017c738f8dcf3e31346de49a390835ffafad3f1b
diff --git a/eli5/lightgbm.py b/eli5/lightgbm.py
index b4510912..c54236b7 100644
--- a/eli5/lightgbm.py
+++ b/eli5/lightgbm.py
@@ -3,8 +3,8 @@
 from collections import defaultdict
 from typing import DefaultDict, Optional
 
-import numpy as np  # type: ignore
-import lightgbm  # type: ignore
+import numpy as np
+import lightgbm
 
 from eli5.explain import explain_weights, explain_prediction
 from eli5._feature_importances import get_feature_importance_explanation
diff --git a/eli5/lightning.py b/eli5/lightning.py
index 417d4f15..2f648064 100644
--- a/eli5/lightning.py
+++ b/eli5/lightning.py
@@ -1,9 +1,9 @@
 # -*- coding: utf-8 -*-
 from __future__ import absolute_import
 
-from lightning.impl.base import BaseEstimator  # type: ignore
-from lightning import classification, regression   # type: ignore
-from sklearn.multiclass import OneVsRestClassifier   # type: ignore
+from lightning.impl.base import BaseEstimator
+from lightning import classification, regression
+from sklearn.multiclass import OneVsRestClassifier
 
 from eli5.base import Explanation
 from eli5.base_utils import singledispatch
diff --git a/eli5/lime/_vectorizer.py b/eli5/lime/_vectorizer.py
index d5168454..5356d6cd 100644
--- a/eli5/lime/_vectorizer.py
+++ b/eli5/lime/_vectorizer.py
@@ -2,8 +2,8 @@
 from __future__ import absolute_import
 from typing import Tuple, Callable, Dict, Optional, List
 
-import numpy as np  # type: ignore
-from sklearn.base import BaseEstimator, TransformerMixin  # type: ignore
+import numpy as np
+from sklearn.base import BaseEstimator, TransformerMixin
 
 from eli5.base import DocWeightedSpans, FeatureWeights
 from eli5.sklearn.text import _get_feature_weights_dict
@@ -13,8 +13,8 @@
 class SingleDocumentVectorizer(BaseEstimator, TransformerMixin):
     """ Fake vectorizer which converts document just to a vector of ones """
 
-    def __init__(self, token_pattern=None):
-        # type: (Optional[str]) -> None
+    def __init__(self, token_pattern):
+        # type: (str) -> None
         self.token_pattern = token_pattern
 
     def fit(self, X, y=None):
diff --git a/eli5/lime/lime.py b/eli5/lime/lime.py
index 924675bb..2968da04 100644
--- a/eli5/lime/lime.py
+++ b/eli5/lime/lime.py
@@ -6,12 +6,12 @@
 from __future__ import absolute_import
 from typing import Any, Callable, Dict, Optional
 
-import numpy as np  # type: ignore
-from sklearn.feature_extraction.text import CountVectorizer  # type: ignore
-from sklearn.linear_model import SGDClassifier  # type: ignore
-from sklearn.model_selection import train_test_split  # type: ignore
-from sklearn.utils import check_random_state  # type: ignore
-from sklearn.base import clone, BaseEstimator  # type: ignore
+import numpy as np
+from sklearn.feature_extraction.text import CountVectorizer
+from sklearn.linear_model import SGDClassifier
+from sklearn.model_selection import train_test_split
+from sklearn.utils import check_random_state
+from sklearn.base import clone, BaseEstimator
 
 import eli5
 from eli5.sklearn.utils import sklearn_version
@@ -148,7 +148,7 @@ def __init__(self,
                  rbf_sigma=None,  # type: float
                  random_state=None,
                  expand_factor=10,  # type: Optional[int]
-                 token_pattern=None,  # type: str
+                 token_pattern=None,  # type: Optional[str]
                  ):
         # type: (...) -> None
         self.n_samples = n_samples
@@ -162,7 +162,7 @@ def __init__(self,
         if char_based is None:
             if token_pattern is None:
                 self.char_based = False  # type: Optional[bool]
-                self.token_pattern = DEFAULT_TOKEN_PATTERN
+                self.token_pattern = DEFAULT_TOKEN_PATTERN  # type: str
             else:
                 self.char_based = None
                 self.token_pattern = token_pattern
diff --git a/eli5/lime/samplers.py b/eli5/lime/samplers.py
index ff72f568..d079b471 100644
--- a/eli5/lime/samplers.py
+++ b/eli5/lime/samplers.py
@@ -5,13 +5,13 @@
 from typing import List, Tuple, Any, Union, Dict, Optional
 import six
 
-import numpy as np  # type: ignore
-from scipy.stats import itemfreq  # type: ignore
-from sklearn.base import BaseEstimator, clone  # type: ignore
-from sklearn.neighbors import KernelDensity  # type: ignore
-from sklearn.metrics import pairwise_distances  # type: ignore
-from sklearn.model_selection import GridSearchCV, KFold  # type: ignore
-from sklearn.utils import check_random_state  # type: ignore
+import numpy as np
+from scipy.stats import itemfreq
+from sklearn.base import BaseEstimator, clone
+from sklearn.neighbors import KernelDensity
+from sklearn.metrics import pairwise_distances
+from sklearn.model_selection import GridSearchCV, KFold
+from sklearn.utils import check_random_state
 
 from eli5.utils import vstack
 from eli5.lime.utils import rbf
diff --git a/eli5/lime/textutils.py b/eli5/lime/textutils.py
index e896f347..98da0428 100644
--- a/eli5/lime/textutils.py
+++ b/eli5/lime/textutils.py
@@ -7,8 +7,8 @@
 import math
 from typing import List, Tuple, Union, Optional
 
-import numpy as np  # type: ignore
-from sklearn.utils import check_random_state  # type: ignore
+import numpy as np
+from sklearn.utils import check_random_state
 
 from eli5.utils import indices_to_bool_mask, vstack
 
diff --git a/eli5/lime/utils.py b/eli5/lime/utils.py
index ee271a65..120dbfbd 100644
--- a/eli5/lime/utils.py
+++ b/eli5/lime/utils.py
@@ -2,12 +2,12 @@
 from __future__ import absolute_import
 from typing import List, Any
 
-import numpy as np  # type: ignore
-from scipy.stats import entropy  # type: ignore
-from sklearn.pipeline import Pipeline  # type: ignore
-from sklearn.utils import check_random_state, issparse  # type: ignore
-from sklearn.utils.metaestimators import if_delegate_has_method  # type: ignore
-from sklearn.utils import shuffle as _shuffle  # type: ignore
+import numpy as np
+from scipy.stats import entropy
+from sklearn.pipeline import Pipeline
+from sklearn.utils import check_random_state, issparse
+from sklearn.utils.metaestimators import if_delegate_has_method
+from sklearn.utils import shuffle as _shuffle
 
 from eli5.utils import vstack
 from eli5.sklearn.utils import sklearn_version
diff --git a/eli5/nn/__init__.py b/eli5/nn/__init__.py
index 7bfe97d5..7422dd60 100644
--- a/eli5/nn/__init__.py
+++ b/eli5/nn/__init__.py
@@ -6,6 +6,6 @@
     compute_weights,
 )
 from .text import (
-    gradcam_text_spans,
+    gradcam_spans,
     resize_1d,
 )
\ No newline at end of file
diff --git a/eli5/nn/gradcam.py b/eli5/nn/gradcam.py
index 34af00b1..ec6af934 100644
--- a/eli5/nn/gradcam.py
+++ b/eli5/nn/gradcam.py
@@ -9,7 +9,6 @@
 """
 
 
-# FIXME: rename functions
 def gradcam_heatmap(activations, grads, relu=True, counterfactual=False):
     # type: (np.ndarray, np.ndarray, bool, bool) -> np.ndarray
     """
@@ -196,4 +195,13 @@ def _validate_classification_target(target, output_shape):
     if not (0 <= target < output_nodes):
         raise ValueError('Prediction target index is ' 
                          'outside the required range [0, {}). ',
-                         'Got {}'.format(output_nodes, target))
\ No newline at end of file
+                         'Got {}'.format(output_nodes, target))
+
+
+def _validate_heatmap(heatmap):
+    # type: (np.ndarray) -> None
+    """Utility function to check that the ``heatmap``
+    argument has the right type."""
+    if not isinstance(heatmap, np.ndarray):
+        raise TypeError('heatmap must be a numpy.ndarray instance. '
+                        'Got: "{}" (type "{}").'.format(heatmap, type(heatmap)))
\ No newline at end of file
diff --git a/eli5/nn/text.py b/eli5/nn/text.py
index 4612a61f..81f69155 100644
--- a/eli5/nn/text.py
+++ b/eli5/nn/text.py
@@ -8,15 +8,18 @@
     WeightedSpans,
     DocWeightedSpans,
 )
+from eli5.nn.gradcam import (
+    _validate_heatmap,
+)
 
 
-def gradcam_text_spans(heatmap, # type: np.ndarray
-                       tokens, # type: Union[np.ndarray, list]
-                       doc, # type: np.ndarray
-                       pad_value=None, # type: Optional[Union[int, float]]
-                       pad_token=None, # type: Optional[str]
-                       interpolation_kind='linear' # type: Union[str, int]
-                       ):
+def gradcam_spans(heatmap, # type: np.ndarray
+                  tokens, # type: Union[np.ndarray, list]
+                  doc, # type: np.ndarray
+                  pad_value=None, # type: Optional[Union[int, float]]
+                  pad_token=None, # type: Optional[str]
+                  interpolation_kind='linear' # type: Union[str, int]
+                  ):
     # type: (...) -> Tuple[Union[np.ndarray, list], np.ndarray, WeightedSpans]
     """
     Create text spans from a Grad-CAM ``heatmap`` imposed over ``tokens``.
@@ -30,9 +33,16 @@ def gradcam_text_spans(heatmap, # type: np.ndarray
 
         **Should be rank 1 (no batch dimension).**
 
+
+        :raises TypeError: if ``heatmap`` is wrong type.
+
     tokens : numpy.ndarray or list
         Tokens that will be highlighted using weights from ``heatmap``.
 
+
+        :raises TypeError: if ``tokens`` is wrong type.
+        :raises ValueError: if ``tokens`` contents are unexpected.
+
     doc: numpy.ndarray
         Original input to the network, from which ``heatmap`` was created.
 
@@ -53,8 +63,15 @@ def gradcam_text_spans(heatmap, # type: np.ndarray
         ``tokens`` and ``heatmap`` optionally cut from padding.
         A :class:`eli5.base.WeightedSpans` object with a weight for each token.
     """
-    # FIXME: might want to do this when formatting the explanation?
+    # We call this before returning the explanation, NOT when formatting the explanation
+    # Because WeightedSpans, etc are attributes of a returned explanation
     # TODO: might want to add validation for heatmap and other arguments?
+    _validate_tokens(tokens)
+    _validate_tokens_value(tokens, doc)
+    if isinstance(tokens, list):
+        # convert to a common data type
+        tokens = np.array(tokens)
+
     length = len(tokens)
     heatmap = resize_1d(heatmap, length, interpolation_kind=interpolation_kind)
 
@@ -62,14 +79,16 @@ def gradcam_text_spans(heatmap, # type: np.ndarray
     if pad_value is not None or pad_token is not None:
         # remove padding
         pad_indices = _find_padding(pad_value=pad_value, pad_token=pad_token, doc=doc, tokens=tokens)
-        # If pad_value is not the actual padding value, behaviour is unknown
+        # If passed padding argument is not the actual padding token/value, behaviour is unknown
         tokens, heatmap = _trim_padding(pad_indices, tokens, heatmap)
+
     document = _construct_document(tokens)
     spans = _build_spans(tokens, heatmap, document)
     weighted_spans = WeightedSpans([
         DocWeightedSpans(document, spans=spans)
-    ]) # why list? - for each vectorized - don't need multiple vectorizers?
-       # multiple highlights? - could do positive and negative expl?
+    ])
+    # why do we have a list of WeightedSpans? One for each vectorizer?
+    # But we do not use multiple vectorizers?
     return tokens, heatmap, weighted_spans
 
 
@@ -87,6 +106,9 @@ def resize_1d(heatmap, length, interpolation_kind='linear'):
     heatmap : numpy.ndarray
         Heatmap to be resized.
 
+
+        :raises TypeError: if ``heatmap`` is wrong type.
+
     length : int
         Required width.
 
@@ -102,6 +124,8 @@ def resize_1d(heatmap, length, interpolation_kind='linear'):
     heatmap : numpy.ndarray
         The heatmap resized.
     """
+    _validate_heatmap(heatmap)
+    _validate_length(length)
     if len(heatmap.shape) == 1 and heatmap.shape[0] == 1:
         # single weight, no batch
         heatmap = heatmap.repeat(length)
@@ -144,7 +168,7 @@ def _build_spans(tokens, # type: Union[np.ndarray, list]
 
 def _construct_document(tokens):
     # type: (Union[list, np.ndarray]) -> str
-    """Create a document string by joining ``tokens``."""
+    """Create a document string by joining ``tokens`` sequence."""
     if _is_character_tokenization(tokens):
         sep = ''
     else:
@@ -154,10 +178,7 @@ def _construct_document(tokens):
 
 def _is_character_tokenization(tokens):
     # type: (Union[list, np.ndarray]) -> bool
-    """
-    Check whether tokenization is character-level
-    (returns True) or word-level (returns False).
-    """
+    """Check whether tokenization is character-level (True) or word-level (False)."""
     return any(' ' in t for t in tokens)
 
 
@@ -178,27 +199,27 @@ def _find_padding(pad_value=None, # type: Union[int, float]
     else:
         raise TypeError('Pass "doc" and "pad_value", '
                         'or "tokens" and "pad_token".')
-    # TODO: warn if indices is empty - passed wrong padding char/value?
 
 
 def _find_padding_values(pad_value, doc):
     # type: (Union[int, float], np.ndarray) -> np.ndarray
     if not isinstance(pad_value, (int, float)):
         raise TypeError('"pad_value" must be int or float. Got "{}"'.format(type(pad_value)))
+    _validate_doc(doc)
     values, indices = np.where(doc == pad_value)
     return indices
 
 
 def _find_padding_tokens(pad_token, tokens):
-    # type: (str, Union[list, np.ndarray]) -> np.ndarray
+    # type: (str, np.ndarray) -> np.ndarray
     if not isinstance(pad_token, str):
         raise TypeError('"pad_token" must be str. Got "{}"'.format(type(pad_token)))
-    indices = [idx for idx, token in enumerate(tokens) if token == pad_token]
-    return np.array(indices)
+    indices = np.where(tokens == pad_token)
+    return indices
 
 
 def _trim_padding(pad_indices, # type: np.ndarray
-                  tokens, # type: Union[list, np.ndarray]
+                  tokens, # type: np.ndarray
                   heatmap, # type: np.ndarray
                   ):
     # type: (...) -> Tuple[Union[list, np.ndarray], np.ndarray]
@@ -212,4 +233,82 @@ def _trim_padding(pad_indices, # type: np.ndarray
         # and we can not detect and raise an error if there is padding in the middle of the text
         tokens = np.delete(tokens, pad_indices)
         heatmap = np.delete(heatmap, pad_indices)
-    return tokens, heatmap
\ No newline at end of file
+    return tokens, heatmap
+
+
+def _validate_doc(doc):
+    if not isinstance(doc, np.ndarray):
+        raise TypeError('"doc" must be an instance of numpy.ndarray. '
+                        'Got "{}" (type "{}")'.format(doc, type(doc)))
+
+
+def _validate_length(length):
+    if not isinstance(length, int):
+        raise TypeError('"length" must be an integer. Got "{}" '
+                        '(type "{}")'.format(length, type(length)))
+    if length < 0:
+        raise ValueError('"length" must be a non-negative integer. '
+                         'Got "{}"'.format(length))
+
+
+# TODO:
+# docs for raises in here
+# coverage tests for new validation
+
+
+# FIXME: break this function up
+def _validate_tokens(tokens):
+    # type: (Union[np.ndarray, list]) -> None
+    """Check that ``tokens`` contains correct items and matches ``doc``."""
+    if not isinstance(tokens, (list, np.ndarray)):
+        # wrong type
+        raise TypeError('"tokens" must be list or numpy.ndarray. '
+                        'Got "{}".'.format(tokens))
+    if len(tokens) == 0:
+        # empty list
+        raise ValueError('"tokens" is empty: {}'.format(tokens))
+
+
+def _validate_tokens_value(tokens, doc):
+    # type: (Union[np.ndarray, list], np.ndarray) -> None
+    doc_batch, doc_len = doc.shape[0], doc.shape[1]
+    an_entry = tokens[0]
+    if isinstance(an_entry, str):
+        # no batch
+        if doc_batch != 1:
+            # doc is batched but tokens is not
+            raise ValueError('If passing "tokens" without batch dimension, '
+                             '"doc" must have batch size = 1.'
+                             'Got "doc" with batch size = %d.' % doc_batch)
+        tokens_len = len(tokens)
+    elif isinstance(an_entry, (list, np.ndarray)):
+        # batched
+        tokens_batch = len(tokens)
+        if tokens_batch != doc_batch:
+            # batch lengths do not match
+            raise ValueError('"tokens" must have same number of samples '
+                             'as in doc batch. Got: "tokens" samples: %d, '
+                             'doc samples: %d' % (tokens_batch, doc_batch))
+
+        a_token = an_entry[0]
+        if not isinstance(a_token, str):
+            # actual contents are not strings
+            raise TypeError('Second axis in "tokens" must contain strings. '
+                            'Found "{}" (type "{}")'.format(a_token, type(a_token)))
+
+        # a way to check that all elements match some condition
+        # https://stackoverflow.com/a/35791116/11555448
+        it = iter(tokens)
+        the_len = len(next(it))
+        if not all(len(l) == the_len for l in it):
+            raise ValueError('"tokens" samples do not all have the same length.')
+        tokens_len = the_len
+    else:
+        raise TypeError('"tokens" must be an array of strings, '
+                        'or an array of string arrays. '
+                        'Got "{}".'.format(tokens))
+
+    if tokens_len != doc_len:
+        raise ValueError('"tokens" and "doc" lengths must match. '
+                         '"tokens" length: "%d". "doc" length: "%d"'
+                         % (tokens_len, doc_len))
\ No newline at end of file
diff --git a/eli5/permutation_importance.py b/eli5/permutation_importance.py
index b5c4a3f0..8ff5d40d 100644
--- a/eli5/permutation_importance.py
+++ b/eli5/permutation_importance.py
@@ -13,8 +13,8 @@
 from __future__ import absolute_import
 from typing import Tuple, List, Callable, Any
 
-import numpy as np  # type: ignore
-from sklearn.utils import check_random_state  # type: ignore
+import numpy as np
+from sklearn.utils import check_random_state
 
 
 def iter_shuffled(X, columns_to_shuffle=None, pre_shuffle=False,
diff --git a/eli5/sklearn/explain_prediction.py b/eli5/sklearn/explain_prediction.py
index 88586f05..18dcc36f 100644
--- a/eli5/sklearn/explain_prediction.py
+++ b/eli5/sklearn/explain_prediction.py
@@ -1,10 +1,10 @@
 # -*- coding: utf-8 -*-
 from functools import partial
 
-import numpy as np  # type: ignore
-import scipy.sparse as sp  # type: ignore
-from sklearn.base import BaseEstimator  # type: ignore
-from sklearn.ensemble import (  # type: ignore
+import numpy as np
+import scipy.sparse as sp
+from sklearn.base import BaseEstimator
+from sklearn.ensemble import (
     ExtraTreesClassifier,
     ExtraTreesRegressor,
     GradientBoostingClassifier,
@@ -12,7 +12,7 @@
     RandomForestClassifier,
     RandomForestRegressor,
 )
-from sklearn.linear_model import (  # type: ignore
+from sklearn.linear_model import (
     ElasticNet,  # includes Lasso, MultiTaskElasticNet, etc.
     ElasticNetCV,
     HuberRegressor,
@@ -34,7 +34,7 @@
     SGDRegressor,
     TheilSenRegressor,
 )
-from sklearn.svm import (  # type: ignore
+from sklearn.svm import (
     LinearSVC,
     LinearSVR,
     SVC,
@@ -43,8 +43,8 @@
     NuSVR,
     OneClassSVM,
 )
-from sklearn.multiclass import OneVsRestClassifier  # type: ignore
-from sklearn.tree import (   # type: ignore
+from sklearn.multiclass import OneVsRestClassifier
+from sklearn.tree import (
     DecisionTreeClassifier,
     DecisionTreeRegressor
 )
diff --git a/eli5/sklearn/explain_weights.py b/eli5/sklearn/explain_weights.py
index 019140a6..49010fb4 100644
--- a/eli5/sklearn/explain_weights.py
+++ b/eli5/sklearn/explain_weights.py
@@ -1,11 +1,11 @@
 # -*- coding: utf-8 -*-
 from __future__ import absolute_import
 
-import numpy as np  # type: ignore
+import numpy as np
 
-from sklearn.base import BaseEstimator, RegressorMixin  # type: ignore
-from sklearn.pipeline import Pipeline  # type: ignore
-from sklearn.linear_model import (   # type: ignore
+from sklearn.base import BaseEstimator, RegressorMixin
+from sklearn.pipeline import Pipeline
+from sklearn.linear_model import (
     ElasticNet,  # includes Lasso, MultiTaskElasticNet, etc.
     ElasticNetCV,
     HuberRegressor,
@@ -27,8 +27,8 @@
     SGDRegressor,
     TheilSenRegressor,
 )
-from sklearn.multiclass import OneVsRestClassifier  # type: ignore
-from sklearn.svm import (  # type: ignore
+from sklearn.multiclass import OneVsRestClassifier
+from sklearn.svm import (
     LinearSVC,
     LinearSVR,
     SVC,
@@ -38,8 +38,8 @@
     OneClassSVM,
 )
 # TODO: see https://github.com/scikit-learn/scikit-learn/pull/2250
-from sklearn.naive_bayes import BernoulliNB, MultinomialNB    # type: ignore
-from sklearn.ensemble import (  # type: ignore
+from sklearn.naive_bayes import BernoulliNB, MultinomialNB
+from sklearn.ensemble import (
     GradientBoostingClassifier,
     GradientBoostingRegressor,
     AdaBoostClassifier,
@@ -49,7 +49,7 @@
     ExtraTreesClassifier,
     ExtraTreesRegressor,
 )
-from sklearn.tree import (  # type: ignore
+from sklearn.tree import (
     DecisionTreeClassifier,
     DecisionTreeRegressor,
 )
diff --git a/eli5/sklearn/permutation_importance.py b/eli5/sklearn/permutation_importance.py
index 30ab3cad..370be8be 100644
--- a/eli5/sklearn/permutation_importance.py
+++ b/eli5/sklearn/permutation_importance.py
@@ -2,23 +2,23 @@
 from functools import partial
 from typing import List
 
-import numpy as np  # type: ignore
-from sklearn.model_selection import check_cv  # type: ignore
-from sklearn.utils.metaestimators import if_delegate_has_method  # type: ignore
-from sklearn.utils import check_array, check_random_state  # type: ignore
-from sklearn.base import (  # type: ignore
+import numpy as np
+from sklearn.model_selection import check_cv
+from sklearn.utils.metaestimators import if_delegate_has_method
+from sklearn.utils import check_array, check_random_state
+from sklearn.base import (
     BaseEstimator,
     MetaEstimatorMixin,
     clone,
     is_classifier
 )
-from sklearn.metrics.scorer import check_scoring  # type: ignore
+from sklearn.metrics.scorer import check_scoring
 
 from eli5.permutation_importance import get_score_importances
 from eli5.sklearn.utils import pandas_available
 
 if pandas_available:
-    import pandas as pd   # type: ignore
+    import pandas as pd
 
 CAVEATS_CV_NONE = """
 Feature importances are computed on the same data as used for training, 
@@ -87,11 +87,13 @@ class PermutationImportance(BaseEstimator, MetaEstimatorMixin):
 
     scoring : string, callable or None, default=None
         Scoring function to use for computing feature importances.
-        A string with scoring name (see scikit-learn docs) or
+        A string with scoring name (see scikit-learn `docs`_) or
         a scorer callable object / function with signature
         ``scorer(estimator, X, y)``.
         If ``None``, the ``score`` method of the estimator is used.
 
+        .. _docs: https://scikit-learn.org/stable/modules/model_evaluation.html#common-cases-predefined-values
+
     n_iter : int, default 5
         Number of random shuffle iterations. Decrease to improve speed,
         increase to get more precise estimates.
@@ -212,8 +214,12 @@ def _cv_scores_importances(self, X, y, groups=None, **fit_params):
         cv = check_cv(self.cv, y, is_classifier(self.estimator))
         feature_importances = []  # type: List
         base_scores = []  # type: List[float]
+        weights = fit_params.pop('sample_weight', None)
+        fold_fit_params = fit_params.copy()
         for train, test in cv.split(X, y, groups):
-            est = clone(self.estimator).fit(X[train], y[train], **fit_params)
+            if weights is not None:
+                fold_fit_params['sample_weight'] = weights[train]
+            est = clone(self.estimator).fit(X[train], y[train], **fold_fit_params)
             score_func = partial(self.scorer_, est)
             _base_score, _importances = self._get_score_importances(
                 score_func, X[test], y[test])
diff --git a/eli5/sklearn/text.py b/eli5/sklearn/text.py
index 57296a6f..fb2748bf 100644
--- a/eli5/sklearn/text.py
+++ b/eli5/sklearn/text.py
@@ -1,8 +1,11 @@
 from __future__ import absolute_import
 from typing import Any, Union, Callable, Dict, List, Optional, Set, Tuple
 
-from sklearn.feature_extraction.text import VectorizerMixin  # type: ignore
-from sklearn.pipeline import FeatureUnion  # type: ignore
+from sklearn.pipeline import FeatureUnion
+try:
+    from sklearn.feature_extraction.text import _VectorizerMixin as VectorizerMixin
+except ImportError:  # Changed in scikit-learn 0.22
+    from sklearn.feature_extraction.text import VectorizerMixin
 
 from eli5.base import (
     DocWeightedSpans, WeightedSpans, FeatureWeights, FeatureWeight,
diff --git a/eli5/sklearn/transform.py b/eli5/sklearn/transform.py
index 2d431f8c..8d79f555 100644
--- a/eli5/sklearn/transform.py
+++ b/eli5/sklearn/transform.py
@@ -1,11 +1,11 @@
 """transform_feature_names implementations for scikit-learn transformers
 """
 
-import numpy as np  # type: ignore
-from sklearn.pipeline import Pipeline, FeatureUnion  # type: ignore
-from sklearn.feature_selection.base import SelectorMixin  # type: ignore
+import numpy as np
+from sklearn.pipeline import Pipeline, FeatureUnion
+from sklearn.feature_selection.base import SelectorMixin
 
-from sklearn.preprocessing import (  # type: ignore
+from sklearn.preprocessing import (
     MinMaxScaler,
     StandardScaler,
     MaxAbsScaler,
@@ -26,7 +26,7 @@ def _select_names(est, in_names=None):
     return [in_names[i] for i in np.flatnonzero(mask)]
 
 try:
-    from sklearn.linear_model import (  # type: ignore
+    from sklearn.linear_model import (
         RandomizedLogisticRegression,
         RandomizedLasso,
     )
diff --git a/eli5/sklearn/treeinspect.py b/eli5/sklearn/treeinspect.py
index 373397bd..d0e9f76a 100644
--- a/eli5/sklearn/treeinspect.py
+++ b/eli5/sklearn/treeinspect.py
@@ -7,8 +7,8 @@
 """
 from __future__ import absolute_import, division
 
-from sklearn.base import ClassifierMixin  # type: ignore
-from sklearn.tree import _tree, export_graphviz  # type: ignore
+from sklearn.base import ClassifierMixin
+from sklearn.tree import _tree, export_graphviz
 
 from eli5.base import TreeInfo, NodeInfo
 
diff --git a/eli5/sklearn/unhashing.py b/eli5/sklearn/unhashing.py
index 64b44d86..f4f79b8d 100644
--- a/eli5/sklearn/unhashing.py
+++ b/eli5/sklearn/unhashing.py
@@ -7,14 +7,14 @@
 from itertools import chain
 from typing import List, Iterable, Any, Dict, Tuple, Union
 
-import numpy as np  # type: ignore
+import numpy as np
 import six
-from sklearn.base import BaseEstimator, TransformerMixin  # type: ignore
-from sklearn.feature_extraction.text import (  # type: ignore
+from sklearn.base import BaseEstimator, TransformerMixin
+from sklearn.feature_extraction.text import (
     HashingVectorizer,
     FeatureHasher,
 )
-from sklearn.pipeline import FeatureUnion  # type: ignore
+from sklearn.pipeline import FeatureUnion
 
 from eli5._feature_names import FeatureNames
 
diff --git a/eli5/sklearn/utils.py b/eli5/sklearn/utils.py
index ba3680aa..286d078a 100644
--- a/eli5/sklearn/utils.py
+++ b/eli5/sklearn/utils.py
@@ -3,9 +3,9 @@
 from distutils.version import LooseVersion
 from typing import Any, Optional, List, Tuple
 
-import numpy as np  # type: ignore
-import scipy.sparse as sp  # type: ignore
-from sklearn.multiclass import OneVsRestClassifier  # type: ignore
+import numpy as np
+import scipy.sparse as sp
+from sklearn.multiclass import OneVsRestClassifier
 
 from eli5.sklearn.unhashing import invert_hashing_and_fit, handle_hashing_vec
 from eli5._feature_names import FeatureNames
@@ -214,7 +214,7 @@ def get_num_features(estimator):
 
 
 try:
-    import pandas as pd  # type: ignore
+    import pandas as pd
     pandas_available = True
 except ImportError:
     pandas_available = False
@@ -277,5 +277,5 @@ def sklearn_version():
     >>> sklearn_version() > '0.17'
     True
     """
-    from sklearn import __version__  # type: ignore
+    from sklearn import __version__
     return LooseVersion(__version__)
diff --git a/eli5/sklearn_crfsuite/explain_weights.py b/eli5/sklearn_crfsuite/explain_weights.py
index e2a9435d..6007efd7 100644
--- a/eli5/sklearn_crfsuite/explain_weights.py
+++ b/eli5/sklearn_crfsuite/explain_weights.py
@@ -1,9 +1,9 @@
 # -*- coding: utf-8 -*-
 from __future__ import absolute_import
 
-import numpy as np  # type: ignore
-from scipy import sparse as sp  # type: ignore
-from sklearn_crfsuite import CRF  # type: ignore
+import numpy as np
+from scipy import sparse as sp
+from sklearn_crfsuite import CRF
 
 from eli5.base import Explanation, TargetExplanation, TransitionFeatureWeights
 from eli5.explain import explain_weights
diff --git a/eli5/utils.py b/eli5/utils.py
index f1ea32c4..e5f669f8 100644
--- a/eli5/utils.py
+++ b/eli5/utils.py
@@ -1,6 +1,6 @@
 # -*- coding: utf-8 -*-
-import numpy as np  # type: ignore
-from scipy import sparse as sp  # type: ignore
+import numpy as np
+from scipy import sparse as sp
 
 
 def argsort_k_largest(x, k):
diff --git a/eli5/xgboost.py b/eli5/xgboost.py
index 86d3ad58..ce6a079a 100644
--- a/eli5/xgboost.py
+++ b/eli5/xgboost.py
@@ -4,9 +4,9 @@
 import re
 from typing import Any, Dict, List, Tuple, Optional, Pattern
 
-import numpy as np  # type: ignore
-import scipy.sparse as sp  # type: ignore
-from xgboost import (  # type: ignore
+import numpy as np
+import scipy.sparse as sp
+from xgboost import (
     XGBClassifier,
     XGBRegressor,
     Booster,
diff --git a/setup.py b/setup.py
index 011fdd28..ec561515 100755
--- a/setup.py
+++ b/setup.py
@@ -37,7 +37,6 @@ def get_long_description():
         'scipy',
         'six',
         'scikit-learn >= 0.18',
-        'typing',
         'graphviz',
         'tabulate>=0.7.7',
     ],
@@ -45,6 +44,7 @@ def get_long_description():
         ":python_version<'3.5.6'": [
             'singledispatch >= 3.4.0.3',
         ],
+        ":python_version<'3.5'": ['typing'],
     },
     classifiers=[
         'Development Status :: 4 - Beta',
diff --git a/tests/test_formatters_image.py b/tests/test_formatters_image.py
index bc395afd..c287062f 100644
--- a/tests/test_formatters_image.py
+++ b/tests/test_formatters_image.py
@@ -15,7 +15,6 @@
     _cap_alpha,
     _overlay_heatmap,
     _validate_image,
-    _validate_heatmap,
 )
 from .utils_image import assert_pixel_by_pixel_equal
 import eli5
@@ -153,12 +152,6 @@ def test_validate_image():
         _validate_image(np.zeros((2, 2, 4,)))
 
 
-def test_validate_heatmap():
-    with pytest.raises(TypeError):
-        # heatmap must be a numpy array, not a Pillow image
-        _validate_heatmap(PIL.Image.new('L', (2, 2,)))
-
-
 def test_format_as_image_notransparency(catdog_rgba):
     # heatmap with full transparency
     expl = Explanation('mock',
diff --git a/tests/test_keras.py b/tests/test_keras.py
index 9ea1c50b..b765abc1 100644
--- a/tests/test_keras.py
+++ b/tests/test_keras.py
@@ -33,13 +33,11 @@
 from eli5.keras.explain_prediction import (
     _validate_model,
     _validate_doc,
-    _validate_tokens,
     _get_layer,
     _autoget_layer_image,
     _autoget_layer_text,
 )
 from eli5.keras.gradcam import (
-    _autoget_target_prediction,
     _calc_gradient,
 )
 
@@ -147,42 +145,6 @@ def test_validate_doc():
         _validate_doc(np.zeros((3, 2, 2, 1)))
 
 
-def test_validate_tokens():
-    _validate_tokens(np.zeros((1, 3)), ['a', 'b', 'c'])
-    _validate_tokens(np.zeros((2, 2)), [['a', 'b'], ['c', 'd']])
-
-
-def test_validate_tokens_invalid():
-    with pytest.raises(TypeError):
-        # should be in a list
-        _validate_tokens(np.zeros((1, 1)), 'a')
-    with pytest.raises(ValueError):
-        # empty list
-        _validate_tokens(np.zeros((1, 1)), [])
-    with pytest.raises(ValueError):
-        # single list but multiple samples in batch
-        _validate_tokens(np.zeros((3, 2)), ['a', 'b'])
-
-    # list doesn't contain strings
-    with pytest.raises(TypeError):
-        _validate_tokens(np.zeros((1, 1)), [0])
-    with pytest.raises(TypeError):
-        _validate_tokens(np.zeros((1, 1)), [[0]])
-
-    with pytest.raises(ValueError):
-        # not enough samples in batched list
-        _validate_tokens(np.zeros((3, 1)), np.array([['a'], ['b']]))
-    with pytest.raises(ValueError):
-        # tokens lengths vary
-        _validate_tokens(np.zeros((2, 2)), [['a', 'b'], ['c']])
-    with pytest.raises(ValueError):
-        # tokens sample lengths do not match
-        _validate_tokens(np.zeros((1, 1)), ['a', 'b'])
-    with pytest.raises(TypeError):
-        # too many axes
-        _validate_tokens(np.zeros((1, 1,)), [[['a']]])
-
-
 def test_explain_prediction_attributes(simple_seq_image, dummy_image):
     expl = eli5.explain_prediction(simple_seq_image, np.zeros((1, 32, 32, 1)))
     assert expl.layer is not None
@@ -225,9 +187,14 @@ def test_calc_gradient(differentiable_model):
 
 def test_calc_gradient_nondifferentiable(nondifferentiable_model):
     with pytest.raises(ValueError):
-        grads = _calc_gradient(nondifferentiable_model.output,
-            [nondifferentiable_model.input])
+        _calc_gradient(nondifferentiable_model.output,
+                       [nondifferentiable_model.input])
+
+
 
+# TODO: test chossing multiple target from multiple maximum values, etc
 
 
-# TODO: test_autoget_target_prediction with multiple maximum values, etc
\ No newline at end of file
+def test_import():
+    # test that package imports without errors
+    import eli5.keras
\ No newline at end of file
diff --git a/tests/test_lightning.py b/tests/test_lightning.py
index 6519d9d5..cebbb7a8 100644
--- a/tests/test_lightning.py
+++ b/tests/test_lightning.py
@@ -73,6 +73,7 @@ def test_explain_weights_regressors(boston_train, reg):
                                               has_bias=False)
 
 
+@pytest.mark.xfail(reason='lightning does not work with pandas dataframes any more')
 @pytest.mark.parametrize(['reg'], _instances(_REGRESSORS)[:2])
 def test_explain_prediction_pandas(reg, boston_train):
     _check_explain_prediction_pandas(reg, boston_train)
diff --git a/tests/test_nn.py b/tests/test_nn.py
new file mode 100644
index 00000000..7dce3b8d
--- /dev/null
+++ b/tests/test_nn.py
@@ -0,0 +1,7 @@
+# -*- coding: utf-8 -*-
+"""Test eli5.nn package"""
+
+
+def test_import():
+    # check that package imports without errors
+    import eli5.nn
\ No newline at end of file
diff --git a/tests/test_nn_gradcam.py b/tests/test_nn_gradcam.py
index a82e3d5b..1516a61c 100644
--- a/tests/test_nn_gradcam.py
+++ b/tests/test_nn_gradcam.py
@@ -3,11 +3,13 @@
 
 import pytest
 import numpy as np
+PIL = pytest.importorskip('PIL')
 
 from eli5.nn.gradcam import (
     gradcam_heatmap,
     _validate_targets,
     _validate_classification_target,
+    _validate_heatmap,
 )
 
 
@@ -64,10 +66,19 @@ def test_validate_targets():
 
 
 # target index must correctly reference one of the nodes in the final layer
-def _validate_classification_target():
+def test_validate_classification_target():
     with pytest.raises(ValueError):
         # one over
         _validate_classification_target(2, (1, 2,))
     with pytest.raises(ValueError):
         # one less
-        _validate_classification_target(-1, (1, 1,))
\ No newline at end of file
+        _validate_classification_target(-1, (1, 1,))
+
+
+def test_validate_heatmap():
+    with pytest.raises(TypeError):
+        # heatmap must be a numpy array, not a Pillow image
+        _validate_heatmap(PIL.Image.new('L', (2, 2,)))
+    with pytest.raises(TypeError):
+        # heatmap must not be a Python list
+        _validate_heatmap([2, 3])
\ No newline at end of file
diff --git a/tests/test_nn_text.py b/tests/test_nn_text.py
index 2a9b2110..1beac020 100644
--- a/tests/test_nn_text.py
+++ b/tests/test_nn_text.py
@@ -5,7 +5,7 @@
 import numpy as np
 
 from eli5.nn.text import (
-    gradcam_text_spans,
+    gradcam_spans,
     resize_1d,
     _build_spans,
     _construct_document,
@@ -13,6 +13,8 @@
     _find_padding_values,
     _find_padding_tokens,
     _trim_padding,
+    _validate_tokens,
+    _validate_tokens_value,
 )
 from eli5.base import (
     WeightedSpans,
@@ -66,11 +68,11 @@ def test_find_padding_values():
 
 
 def test_find_padding_tokens():
-    indices = _find_padding_tokens('<PAD>', ['the', 'test', '<PAD>', '<PAD>'])
+    indices = _find_padding_tokens('<PAD>', np.array(['the', 'test', '<PAD>', '<PAD>']))
     np.array_equal(indices, np.array([2, 3]))
 
     with pytest.raises(TypeError):
-        _find_padding_tokens(0, ['<PAD>'])
+        _find_padding_tokens(0, np.array(['<PAD>']))
 
 
 @pytest.mark.parametrize('pad_indices, tokens, heatmap, expected_tokens, expected_heatmap', [
@@ -93,16 +95,48 @@ def test_trim_padding_invalid():
     assert np.array_equal(tokens, tokens_trimmed)
     assert np.array_equal(heatmap, heatmap_trimmed)
 
-    # with pytest.raises(ValueError):
-    #     _trim_padding([1], ['a', 'PAD', 'b'], np.array([1, 0, 2]))
 
-
-def test_gradcam_text_spans():
-    heatmap, tokens, doc = np.array([2.0]), ['a'], [2]
-    res_tokens, res_heatmap, res_weighted_spans = gradcam_text_spans(heatmap, tokens, doc)
+def test_gradcam_spans():
+    heatmap, tokens, doc = np.array([2.0]), ['a'], np.array([[2]])
+    res_tokens, res_heatmap, res_weighted_spans = gradcam_spans(heatmap, tokens, doc)
     assert np.array_equal(heatmap, res_heatmap)
     assert np.array_equal(tokens, res_tokens)
     assert res_weighted_spans == WeightedSpans([DocWeightedSpans(
                                                 'a',
                                                 spans=[('a', [(0, 1)], 2.0)]
-                                                )])
\ No newline at end of file
+                                                )])
+
+
+def test_validate_tokens_invalid():
+    # should be in a list or numpy array
+    with pytest.raises(TypeError):
+        _validate_tokens('a')
+
+    # empty list
+    with pytest.raises(ValueError):
+        _validate_tokens([])
+
+
+def test_validate_tokens_value_invalid():
+    with pytest.raises(ValueError):
+        # single tokens list but multiple samples in input doc
+        _validate_tokens_value(['a', 'b'], np.zeros((3, 2)))
+
+    # list doesn't contain strings
+    with pytest.raises(TypeError):
+        _validate_tokens_value([0], np.zeros((1, 1)))
+    with pytest.raises(TypeError):
+        _validate_tokens_value([[0]], np.zeros((1, 1)))
+
+    with pytest.raises(ValueError):
+        # not enough samples in batched list
+        _validate_tokens_value(np.array([['a'], ['b']]), np.zeros((3, 1)))
+    with pytest.raises(ValueError):
+        # tokens lengths vary
+        _validate_tokens_value([['a', 'b'], ['c']], np.zeros((2, 2)))
+    with pytest.raises(ValueError):
+        # tokens sample lengths do not match
+        _validate_tokens_value(['a', 'b'], np.zeros((1, 1)))
+    with pytest.raises(TypeError):
+        # too many axes
+        _validate_tokens_value([[['a']]], np.zeros((1, 1,)))
\ No newline at end of file
diff --git a/tests/test_permutation_importance.py b/tests/test_permutation_importance.py
index effb4ff9..f4b95233 100644
--- a/tests/test_permutation_importance.py
+++ b/tests/test_permutation_importance.py
@@ -41,7 +41,7 @@ def is_shuffled(X, X_sh, col):
 
 def test_get_feature_importances(boston_train):
     X, y, feat_names = boston_train
-    svr = SVR(C=20).fit(X, y)
+    svr = SVR(C=20, gamma='auto').fit(X, y)
     score, importances = get_score_importances(svr.score, X, y)
     assert score > 0.7
     importances = dict(zip(feat_names, np.mean(importances, axis=0)))
diff --git a/tests/test_sklearn_explain_prediction.py b/tests/test_sklearn_explain_prediction.py
index 646c8c7d..277fee46 100644
--- a/tests/test_sklearn_explain_prediction.py
+++ b/tests/test_sklearn_explain_prediction.py
@@ -379,7 +379,7 @@ def test_explain_linear_binary(newsgroups_train_binary, clf):
 
 def test_explain_one_class_svm():
     X = np.array([[0, 0], [0, 1], [5, 3], [93, 94], [90, 91]])
-    clf = OneClassSVM(kernel='linear', random_state=42).fit(X)
+    clf = OneClassSVM(kernel='linear').fit(X)
     res = explain_prediction(clf, X[0])
     assert res.targets[0].score < 0
     for expl in format_as_all(res, clf):
diff --git a/tests/test_sklearn_explain_weights.py b/tests/test_sklearn_explain_weights.py
index 6ca5f519..7f5469c2 100644
--- a/tests/test_sklearn_explain_weights.py
+++ b/tests/test_sklearn_explain_weights.py
@@ -210,7 +210,7 @@ def test_explain_linear_unsupported_multiclass(clf, newsgroups_train):
 
 def test_explain_one_class_svm():
     X = np.array([[0,0], [0, 1], [5, 3], [93, 94], [90, 91]])
-    clf = OneClassSVM(kernel='linear', random_state=42).fit(X)
+    clf = OneClassSVM(kernel='linear', gamma='auto').fit(X)
     res = explain_weights(clf)
     assert len(res.targets) == 1
     target = res.targets[0]
@@ -451,7 +451,7 @@ def test_explain_random_forest_and_tree_feature_filter(newsgroups_train, clf):
 
 
 def test_explain_empty(newsgroups_train):
-    clf = LogisticRegression(C=0.01, penalty='l1', random_state=42)
+    clf = LogisticRegression(C=0.01, penalty='l1', solver='liblinear', random_state=42)
     docs, y, target_names = newsgroups_train
     vec = TfidfVectorizer()
 
diff --git a/tests/test_sklearn_permutation_importance.py b/tests/test_sklearn_permutation_importance.py
index 19e54e2d..4ffec3ba 100644
--- a/tests/test_sklearn_permutation_importance.py
+++ b/tests/test_sklearn_permutation_importance.py
@@ -3,7 +3,7 @@
 import numpy as np
 from sklearn.base import is_classifier, is_regressor
 from sklearn.svm import SVR, SVC
-from sklearn.ensemble import RandomForestRegressor
+from sklearn.ensemble import RandomForestRegressor, RandomForestClassifier
 from sklearn.model_selection import train_test_split, cross_val_score
 from sklearn.pipeline import make_pipeline
 from sklearn.feature_selection import SelectFromModel
@@ -73,7 +73,7 @@ def test_cv(boston_train):
         *boston_train, noise_ratio=0.99)
 
     reg = PermutationImportance(
-        SVR(C=100),
+        SVR(C=100, gamma='auto'),
         random_state=42,
         cv=None,
         n_iter=50,  # use the same number of experiments as with cv=10
@@ -86,7 +86,7 @@ def test_cv(boston_train):
 
     # CV feature importances
     reg = PermutationImportance(
-        SVR(C=100),
+        SVR(C=100, gamma='auto'),
         random_state=42,
         cv=10,
     ).fit(X_test, y_test)
@@ -132,9 +132,9 @@ def test_feature_selection(boston_train):
         ),
         threshold=0.1,
     )
-    pipe = make_pipeline(sel, SVR(C=10))
+    pipe = make_pipeline(sel, SVR(C=10, gamma='auto'))
     score1 = cross_val_score(pipe, X, y).mean()
-    score2 = cross_val_score(SVR(C=10), X, y).mean()
+    score2 = cross_val_score(SVR(C=10, gamma='auto'), X, y).mean()
     print(score1, score2)
     assert score1 > score2
 
@@ -165,6 +165,7 @@ def test_explain_weights(iris_train):
         for _expl in res:
             assert "petal width (cm)" in _expl
 
+
 def test_pandas_xgboost_support(iris_train):
     xgboost = pytest.importorskip('xgboost')
     pd = pytest.importorskip('pandas')
@@ -175,3 +176,17 @@ def test_pandas_xgboost_support(iris_train):
     est.fit(X, y)
     # we expect no exception to be raised here when using xgboost with pd.DataFrame
     perm = PermutationImportance(est).fit(X, y) 
+
+
+def test_cv_sample_weight(iris_train):
+    X, y, feature_names, target_names = iris_train
+    weights_ones = np.ones(len(y))
+    model = RandomForestClassifier(random_state=42)
+
+    # we expect no exception to be raised when passing weights with a CV
+    perm_weights = PermutationImportance(model, cv=5, random_state=42).\
+        fit(X, y, sample_weight=weights_ones)
+    perm = PermutationImportance(model, cv=5, random_state=42).fit(X, y)
+
+    # passing a vector of weights filled with one should be the same as passing no weights
+    assert (perm.feature_importances_ == perm_weights.feature_importances_).all()
\ No newline at end of file
diff --git a/tests/test_sklearn_transform.py b/tests/test_sklearn_transform.py
index aa200937..0c7123f5 100644
--- a/tests/test_sklearn_transform.py
+++ b/tests/test_sklearn_transform.py
@@ -81,19 +81,19 @@ def selection_score_func(X, y):
     (VarianceThreshold(1.0), ['<NAME2>']),
     (GenericUnivariateSelect(), ['<NAME2>']),
     (GenericUnivariateSelect(mode='k_best', param=2), ['<NAME2>', '<NAME3>']),
-    (SelectFromModel(LogisticRegression('l1', C=0.01, random_state=42)),
+    (SelectFromModel(LogisticRegression('l1', C=0.01, solver='liblinear', random_state=42, multi_class='ovr')),
      ['<NAME0>', '<NAME2>']),
     (SelectFromModel(
         PermutationImportance(
-            LogisticRegression(random_state=42),
+            LogisticRegression(solver='liblinear', random_state=42),
             cv=5, random_state=42, refit=False,
         ),
         threshold=0.1,
      ),
      ['<NAME2>', '<NAME3>']),
-    (RFE(LogisticRegression(random_state=42), 2),
+    (RFE(LogisticRegression(solver='liblinear', random_state=42, multi_class='ovr'), 2),
      ['<NAME1>', '<NAME3>']),
-    (RFECV(LogisticRegression(random_state=42)),
+    (RFECV(LogisticRegression(solver='liblinear', random_state=42, multi_class='ovr'), cv=3),
      ['<NAME0>', '<NAME1>', '<NAME2>', '<NAME3>']),
 ] + _additional_test_cases)
 def test_transform_feature_names_iris(transformer, expected, iris_train):
diff --git a/tox.ini b/tox.ini
index 81465eba..919c0eb2 100644
--- a/tox.ini
+++ b/tox.ini
@@ -87,10 +87,10 @@ commands={[testenv:py35-extra]commands}
 basepython=python3.6
 deps=
     {[testenv]deps}
-    mypy == 0.641
+    mypy == 0.750
     lxml
 commands=
-    mypy --html-report ./mypy-cov --check-untyped-defs eli5
+    mypy --html-report ./mypy-cov --check-untyped-defs --ignore-missing-imports eli5
 
 
 [testenv:docs]