Week 9 - Dataset preprocessing

Before we utilize machine learning algorithms we must first prepare our dataset. This can often take a significant amount of time and can have a large impact on the performance of our models.

We will be looking at four different types of data:

  • Tabular data
  • Image data
  • Text

Tabular data

We will look at three different steps we may need to take when handling tabular data:

  • Missing data
  • Normalization
  • Categorical data

Image data

Image data can present a number of issues that we must address to maximize performance:

  • Histogram normalization
  • Windows
  • Pyramids (for detection at different scales)
  • Centering

Text

Text can present a number of issues, mainly due to the number of words that can be found in our features. There are a number of ways we can convert from text to usable features:

  • Bag of words
  • Parsing

In [68]:
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

%matplotlib inline

Tabular data

  • Missing data
  • Normalization
  • Categorical data

Missing data

There are a number of ways to handle missing data:

  • Drop all records with a value missing
  • Substitute all missing values with an average value
  • Substitute all missing values with some placeholder value, i.e. 0, 1e9, -1e9, etc
  • Predict missing values based on other attributes
  • Add additional feature indicating when a value is missing

If the machine learning model will be used with new data it is important to consider the possibility of receiving records with values missing that we have not observed previously in the training dataset.

The simplest approach is to remove any records that have missing data. Unfortunately missing values are often not randomly distributed through a dataset and removing them can introduce bias.

An alternative approach is to substitute the missing values. This can be with the mean of the feature across all the records or the value can be predicted based on the values of the other features in the dataset. Placeholder values can also be used with decision trees but do not work as well for most other algorithms.

Finally, missing values can themselves be useful features. Adding an additional feature indicating when a value is missing is often used to include this information.


In [69]:
from sklearn import linear_model

x = np.array([[0, 0], [1, 1], [2, 2]])
y = np.array([0, 1, 2])
print(x,y)

clf = linear_model.LinearRegression()
clf.fit(x, y)
print(clf.coef_)

x_missing = np.array([[0, 0], [1, np.nan], [2, 2]])
print(x_missing, y)

clf = linear_model.LinearRegression()
clf.fit(x_missing, y)
print(clf.coef_)


[[0 0]
 [1 1]
 [2 2]] [0 1 2]
[ 0.5  0.5]
[[  0.   0.]
 [  1.  nan]
 [  2.   2.]] [0 1 2]
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-69-e50b924921a0> in <module>()
     13 
     14 clf = linear_model.LinearRegression()
---> 15 clf.fit(x_missing, y)
     16 print(clf.coef_)

C:\Users\stree\Anaconda3\lib\site-packages\sklearn\linear_model\base.py in fit(self, X, y, sample_weight)
    425         n_jobs_ = self.n_jobs
    426         X, y = check_X_y(X, y, accept_sparse=['csr', 'csc', 'coo'],
--> 427                          y_numeric=True, multi_output=True)
    428 
    429         if ((sample_weight is not None) and np.atleast_1d(sample_weight).ndim > 1):

C:\Users\stree\Anaconda3\lib\site-packages\sklearn\utils\validation.py in check_X_y(X, y, accept_sparse, dtype, order, copy, force_all_finite, ensure_2d, allow_nd, multi_output, ensure_min_samples, ensure_min_features, y_numeric, warn_on_dtype, estimator)
    508     X = check_array(X, accept_sparse, dtype, order, copy, force_all_finite,
    509                     ensure_2d, allow_nd, ensure_min_samples,
--> 510                     ensure_min_features, warn_on_dtype, estimator)
    511     if multi_output:
    512         y = check_array(y, 'csr', force_all_finite=True, ensure_2d=False,

C:\Users\stree\Anaconda3\lib\site-packages\sklearn\utils\validation.py in check_array(array, accept_sparse, dtype, order, copy, force_all_finite, ensure_2d, allow_nd, ensure_min_samples, ensure_min_features, warn_on_dtype, estimator)
    396                              % (array.ndim, estimator_name))
    397         if force_all_finite:
--> 398             _assert_all_finite(array)
    399 
    400     shape_repr = _shape_repr(array.shape)

C:\Users\stree\Anaconda3\lib\site-packages\sklearn\utils\validation.py in _assert_all_finite(X)
     52             and not np.isfinite(X).all()):
     53         raise ValueError("Input contains NaN, infinity"
---> 54                          " or a value too large for %r." % X.dtype)
     55 
     56 

ValueError: Input contains NaN, infinity or a value too large for dtype('float64').

In [71]:
import pandas as pd

x = pd.DataFrame([[0,1,2,3,4,5,6],
                  [2,np.nan,7,4,9,1,3],
                  [0.1,0.12,0.11,0.15,0.16,0.11,0.14],
                  [100,120,np.nan,127,130,121,124],
                  [4,1,7,9,0,2,np.nan]], ).T
x.columns =['A', 'B', 'C', 'D', 'E']

y = pd.Series([29.0,
 31.2,
 63.25,
 57.27,
 66.3,
 26.21,
 48.24])

print(x, y)


   A   B     C    D   E
0  0   2  0.10  100   4
1  1 NaN  0.12  120   1
2  2   7  0.11  NaN   7
3  3   4  0.15  127   9
4  4   9  0.16  130   0
5  5   1  0.11  121   2
6  6   3  0.14  124 NaN 0    29.00
1    31.20
2    63.25
3    57.27
4    66.30
5    26.21
6    48.24
dtype: float64

In [7]:
x.dropna()


Out[7]:
A B C D E
0 0 2 0.10 100 4
3 3 4 0.15 127 9
4 4 9 0.16 130 0
5 5 1 0.11 121 2

In [8]:
x.fillna(value={'A':1000,'B':2000,'C':3000,'D':4000,'E':5000})


Out[8]:
A B C D E
0 0 2 0.10 100 4
1 1 2000 0.12 120 1
2 2 7 0.11 4000 7
3 3 4 0.15 127 9
4 4 9 0.16 130 0
5 5 1 0.11 121 2
6 6 3 0.14 124 5000

In [9]:
x.fillna(value=x.mean())


Out[9]:
A B C D E
0 0 2.000000 0.10 100.000000 4.000000
1 1 4.333333 0.12 120.000000 1.000000
2 2 7.000000 0.11 120.333333 7.000000
3 3 4.000000 0.15 127.000000 9.000000
4 4 9.000000 0.16 130.000000 0.000000
5 5 1.000000 0.11 121.000000 2.000000
6 6 3.000000 0.14 124.000000 3.833333

Normalization

Many machine learning algorithms expect features to have similar distributions and scales.

A classic example is gradient descent, if features are on different scales some weights will update faster than others because the feature values scale the weight updates.

There are two common approaches to normalization:

  • Z-score standardization
  • Min-max scaling

Z-score standardization

Z-score standardization rescales values so that they have a mean of zero and a standard deviation of 1. Specifically we perform the following transformation:

$$z = \frac{x - \mu}{\sigma}$$

Min-max scaling

An alternative is min-max scaling that transforms data into the range of 0 to 1. Specifically:

$$x_{norm} = \frac{x - x_{min}}{x_{max} - x_{min}}$$

Min-max scaling is less commonly used but can be useful for image data and in some neural networks.


In [12]:
x_filled = x.fillna(value=x.mean())

print(x_filled)


   A         B     C           D         E
0  0  2.000000  0.10  100.000000  4.000000
1  1  4.333333  0.12  120.000000  1.000000
2  2  7.000000  0.11  120.333333  7.000000
3  3  4.000000  0.15  127.000000  9.000000
4  4  9.000000  0.16  130.000000  0.000000
5  5  1.000000  0.11  121.000000  2.000000
6  6  3.000000  0.14  124.000000  3.833333

In [13]:
x_norm = (x_filled - x_filled.min()) / (x_filled.max() - x_filled.min())

print(x_norm)


          A         B         C         D         E
0  0.000000  0.125000  0.000000  0.000000  0.444444
1  0.166667  0.416667  0.333333  0.666667  0.111111
2  0.333333  0.750000  0.166667  0.677778  0.777778
3  0.500000  0.375000  0.833333  0.900000  1.000000
4  0.666667  1.000000  1.000000  1.000000  0.000000
5  0.833333  0.000000  0.166667  0.700000  0.222222
6  1.000000  0.250000  0.666667  0.800000  0.425926

In [14]:
from sklearn import preprocessing

scaling = preprocessing.MinMaxScaler().fit(x_filled)
scaling.transform(x_filled)


Out[14]:
array([[ 0.        ,  0.125     ,  0.        ,  0.        ,  0.44444444],
       [ 0.16666667,  0.41666667,  0.33333333,  0.66666667,  0.11111111],
       [ 0.33333333,  0.75      ,  0.16666667,  0.67777778,  0.77777778],
       [ 0.5       ,  0.375     ,  0.83333333,  0.9       ,  1.        ],
       [ 0.66666667,  1.        ,  1.        ,  1.        ,  0.        ],
       [ 0.83333333,  0.        ,  0.16666667,  0.7       ,  0.22222222],
       [ 1.        ,  0.25      ,  0.66666667,  0.8       ,  0.42592593]])

In [ ]:


In [ ]:

Categorical data

Categorical data can take one of a number of possible values. The different categories may be related to each other or be largely independent and unordered.

Continuous variables can be converted to categorical variables by applying a threshold.


In [15]:
x = pd.DataFrame([[0,1,2,3,4,5,6],
                  [2,np.nan,7,4,9,1,3],
                  [0.1,0.12,0.11,0.15,0.16,0.11,0.14],
                  [100,120,np.nan,127,130,121,124],
                  ['Green','Red','Blue','Blue','Green','Red','Green']], ).T
x.columns = index=['A', 'B', 'C', 'D', 'E']

print(x)


   A    B     C    D      E
0  0    2   0.1  100  Green
1  1  NaN  0.12  120    Red
2  2    7  0.11  NaN   Blue
3  3    4  0.15  127   Blue
4  4    9  0.16  130  Green
5  5    1  0.11  121    Red
6  6    3  0.14  124  Green

In [16]:
x_cat = x.copy()

for val in x['E'].unique():
    x_cat['E_{0}'.format(val)] = x_cat['E'] == val

x_cat


Out[16]:
A B C D E E_Green E_Red E_Blue
0 0 2 0.1 100 Green True False False
1 1 NaN 0.12 120 Red False True False
2 2 7 0.11 NaN Blue False False True
3 3 4 0.15 127 Blue False False True
4 4 9 0.16 130 Green True False False
5 5 1 0.11 121 Red False True False
6 6 3 0.14 124 Green True False False

In [ ]:

Exercises

  1. Substitute missing values in x with the column mean and add an additional column to indicate when missing values have been substituted. The isnull method on the pandas dataframe may be useful.
  2. Convert x to the z-scaled values. The StandardScaler method in the preprocessing module can be used or the z-scaled values calculated directly.
  3. Convert x['C'] into a categorical variable using a threshold of 0.125

In [72]:
x, x.isnull()


Out[72]:
(   A   B     C    D   E
 0  0   2  0.10  100   4
 1  1 NaN  0.12  120   1
 2  2   7  0.11  NaN   7
 3  3   4  0.15  127   9
 4  4   9  0.16  130   0
 5  5   1  0.11  121   2
 6  6   3  0.14  124 NaN,        A      B      C      D      E
 0  False  False  False  False  False
 1  False   True  False  False  False
 2  False  False  False   True  False
 3  False  False  False  False  False
 4  False  False  False  False  False
 5  False  False  False  False  False
 6  False  False  False  False   True)

In [73]:
x['B_isnull'] = x['B'].isnull()
x


Out[73]:
A B C D E B_isnull
0 0 2 0.10 100 4 False
1 1 NaN 0.12 120 1 True
2 2 7 0.11 NaN 7 False
3 3 4 0.15 127 9 False
4 4 9 0.16 130 0 False
5 5 1 0.11 121 2 False
6 6 3 0.14 124 NaN False

In [78]:
(x[['A', 'B', 'C', 'D', 'E']] - x[['A', 'B', 'C', 'D', 'E']].mean()) / \
                                            x[['A', 'B', 'C', 'D', 'E']].std()


Out[78]:
A B C D E
0 -1.38873 -0.758365 -1.185957 -1.912235 0.047015
1 -0.92582 NaN -0.312094 -0.031348 -0.799259
2 -0.46291 0.866703 -0.749025 NaN 0.893290
3 0.00000 -0.108338 0.998700 0.626962 1.457473
4 0.46291 1.516730 1.435632 0.909095 -1.081351
5 0.92582 -1.083378 -0.749025 0.062696 -0.517168
6 1.38873 -0.433351 0.561769 0.344829 NaN

In [75]:
x_scaled = _74

In [76]:
x_scaled.mean(), x_scaled.std()


Out[76]:
(A    3.172066e-17
 B    5.551115e-17
 C    1.586033e-16
 D    4.440892e-16
 E   -5.551115e-17
 dtype: float64, A    1
 B    1
 C    1
 D    1
 E    1
 dtype: float64)

In [79]:
x['C_cat'] = x['C'] > 0.125

In [80]:
x


Out[80]:
A B C D E B_isnull C_cat
0 0 2 0.10 100 4 False False
1 1 NaN 0.12 120 1 True False
2 2 7 0.11 NaN 7 False False
3 3 4 0.15 127 9 False True
4 4 9 0.16 130 0 False True
5 5 1 0.11 121 2 False False
6 6 3 0.14 124 NaN False True

Image data

Depending on the type of task being performed there are a variety of steps we may want to take in working with images:

  • Histogram normalization
  • Windows and pyramids (for detection at different scales)
  • Centering

Occasionally the camera used to generate an image will use 10- to 14-bits while a 16-bit file format will be used. In this situation all the pixel intensities will be in the lower values. Rescaling to the full range (or to 0-1) can be useful.

Further processing can be done to alter the histogram of the image.

When looking for particular features in an image a sliding window can be used to check different locations. This can be combined with an image pyramid to detect features at different scales. This is often needed when objects can be at different distances from the camera.

If objects are sparsely distributed in an image a faster approach than using sliding windows is to identify objects with a simple threshold and then test only the bounding boxes containing objects. Before running these through a model centering based on intensity can be a useful approach. Small offsets, rotations and skewing can be used to generate additional training data.


In [81]:
# http://scikit-image.org/docs/stable/auto_examples/color_exposure/plot_equalize.html#example-color-exposure-plot-equalize-py

import matplotlib
import matplotlib.pyplot as plt
import numpy as np

from skimage import data, img_as_float
from skimage import exposure


matplotlib.rcParams['font.size'] = 8


def plot_img_and_hist(img, axes, bins=256):
    """Plot an image along with its histogram and cumulative histogram.

    """
    img = img_as_float(img)
    ax_img, ax_hist = axes
    ax_cdf = ax_hist.twinx()

    # Display image
    ax_img.imshow(img, cmap=plt.cm.gray)
    ax_img.set_axis_off()
    ax_img.set_adjustable('box-forced')

    # Display histogram
    ax_hist.hist(img.ravel(), bins=bins, histtype='step', color='black')
    ax_hist.ticklabel_format(axis='y', style='scientific', scilimits=(0, 0))
    ax_hist.set_xlabel('Pixel intensity')
    ax_hist.set_xlim(0, 1)
    ax_hist.set_yticks([])

    # Display cumulative distribution
    img_cdf, bins = exposure.cumulative_distribution(img, bins)
    ax_cdf.plot(bins, img_cdf, 'r')
    ax_cdf.set_yticks([])

    return ax_img, ax_hist, ax_cdf


# Load an example image
img = data.moon()

# Contrast stretching
p2, p98 = np.percentile(img, (2, 98))
img_rescale = exposure.rescale_intensity(img, in_range=(p2, p98))

# Equalization
img_eq = exposure.equalize_hist(img)

# Adaptive Equalization
img_adapteq = exposure.equalize_adapthist(img, clip_limit=0.03)

# Display results
fig = plt.figure(figsize=(8, 5))
axes = np.zeros((2,4), dtype=np.object)
axes[0,0] = fig.add_subplot(2, 4, 1)
for i in range(1,4):
    axes[0,i] = fig.add_subplot(2, 4, 1+i, sharex=axes[0,0], sharey=axes[0,0])
for i in range(0,4):
    axes[1,i] = fig.add_subplot(2, 4, 5+i)

ax_img, ax_hist, ax_cdf = plot_img_and_hist(img, axes[:, 0])
ax_img.set_title('Low contrast image')

y_min, y_max = ax_hist.get_ylim()
ax_hist.set_ylabel('Number of pixels')
ax_hist.set_yticks(np.linspace(0, y_max, 5))

ax_img, ax_hist, ax_cdf = plot_img_and_hist(img_rescale, axes[:, 1])
ax_img.set_title('Contrast stretching')

ax_img, ax_hist, ax_cdf = plot_img_and_hist(img_eq, axes[:, 2])
ax_img.set_title('Histogram equalization')

ax_img, ax_hist, ax_cdf = plot_img_and_hist(img_adapteq, axes[:, 3])
ax_img.set_title('Adaptive equalization')

ax_cdf.set_ylabel('Fraction of total intensity')
ax_cdf.set_yticks(np.linspace(0, 1, 5))

# prevent overlap of y-axis labels
fig.tight_layout()
plt.show()


C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 13
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 13
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 37
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 37
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 5
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 5
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 128
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 128
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 52
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 52
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 32
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 32
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 51
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 51
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 64
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 64
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 85
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 85
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 26
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 26
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 9
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 9
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 86
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 86
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 3
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 3
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 22
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 22
  n_excess -= hist[under].size

In [66]:
from sklearn.feature_extraction import image

img = data.page()
fig, ax = plt.subplots(1,1)
ax.imshow(img, cmap=plt.cm.gray)
ax.set_axis_off()
plt.show()
print(img.shape)


(191, 384)

In [67]:
patches = image.extract_patches_2d(img, (20, 20), max_patches=2, random_state=0)
patches.shape
plt.imshow(patches[0], cmap=plt.cm.gray)
plt.show()



In [23]:
from sklearn import datasets
digits = datasets.load_digits()
#print(digits.DESCR)
fig, ax = plt.subplots(1,1, figsize=(1,1))
ax.imshow(digits.data[0].reshape((8,8)), cmap=plt.cm.gray, interpolation='nearest')


Out[23]:
<matplotlib.image.AxesImage at 0x15213139128>

In [ ]:

Text

When working with text the simplest approach is known as bag of words. In this approach we simply count the number of instances of each word, and then adjust the values based on how commonly the word is used.

The first task is to break a piece of text up into individual tokens. The number of occurrences of each word is then recorded. More rarely used words are likely to be more interesting and so word counts are scaled by the inverse document frequency.

We can extend this to look at not just individual words but also bigrams and trigrams.


In [82]:
from sklearn.datasets import fetch_20newsgroups
twenty_train = fetch_20newsgroups(subset='train',
    categories=['comp.graphics', 'sci.med'], shuffle=True, random_state=0)

print(twenty_train.target_names)

from sklearn.feature_extraction.text import CountVectorizer
count_vect = CountVectorizer()
X_train_counts = count_vect.fit_transform(twenty_train.data)
print(X_train_counts.shape)

from sklearn.feature_extraction.text import TfidfTransformer
tfidf_transformer = TfidfTransformer()
X_train_tfidf = tfidf_transformer.fit_transform(X_train_counts)
print(X_train_tfidf.shape, X_train_tfidf[:5,:15].toarray())


['comp.graphics', 'sci.med']
(1178, 24614)
(1178, 24614) [[ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.]]

In [83]:
print(twenty_train.data[0])


From: harti@mikro.ee.tu-berlin.de (Stefan Hartmann (Behse))
Subject: Genoa graphics board Drivers FTP site!
Article-I.D.: mailgzrz.1qpf1r$9ti
Organization: TUBerlin/ZRZ
Lines: 29
NNTP-Posting-Host: mikro.ee.tu-berlin.de

Hi,

well I have opened up a FTP site for getting the latest software drivers
for Genoa graphics cards.

Here is how to access it:

ftp 192.109.42.11
login:ftp
password:ftp
cd pub/genoa
ls -l
binary
prompt
hash

(now if you wanna have the latest drivers for the 7900 board)

cd 7000series
mget *

quit

This is the sequence to get the drivers.

If you have any further question, please email me.

Best regards, Stefan Hartmann
email to: harti@mikro.ee.tu-berlin.de


In [84]:
count_vect = CountVectorizer()
X_train_counts = count_vect.fit_transform(twenty_train.data[0:1])
print(X_train_counts[0].toarray())
print(count_vect.vocabulary_.keys())


[[1 1 1 1 1 1 1 1 1 1 1 1 1 3 1 1 2 1 2 3 4 3 2 3 1 5 1 3 1 1 2 2 2 1 3 1 1
  1 1 2 2 1 2 1 1 1 1 1 1 3 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 2 1 5 1 3 3 1 1 1
  1 2 1]]
dict_keys(['stefan', 'getting', 'is', 'best', 'ftp', 'mailgzrz', 'quit', 'wanna', 'now', 'subject', 'organization', 'how', '7900', 'up', 'further', '192', 'mget', 'from', 'any', 'please', 'have', 'graphics', 'question', 'the', 'me', 'here', 'site', '9ti', 'genoa', 'access', 'berlin', 'for', 'de', 'if', 'sequence', 'latest', 'ls', 'article', '109', 'hash', 'board', 'opened', 'password', 'email', '7000series', 'tuberlin', 'behse', 'to', 'mikro', 'tu', 'ee', 'software', 'posting', 'it', 'you', 'drivers', '1qpf1r', 'lines', 'hartmann', 'zrz', '42', 'host', 'cd', 'well', 'pub', '29', 'nntp', 'cards', 'this', 'hi', '11', 'binary', 'get', 'regards', 'login', 'prompt', 'harti'])

Exercises

  1. Choose one of the histogram processing methods and apply it to the page example.
  2. Take patches for the page example used above at different scales (10, 20 and 40 pixels). The resulting patches should be rescaled to have the same size.
  3. Change the vectorization approach to ignore very common words such as 'the' and 'a'. These are known as stop words. Reading the documentation should help.
  4. Change the vectorization approach to consider both single words and sequences of 2 words. Reading the documentation should help.

In [87]:
from sklearn.feature_extraction import image

img = data.page()
fig, ax = plt.subplots(1,1)
ax.imshow(img, cmap=plt.cm.gray)
ax.set_axis_off()
plt.show()
print(img.shape)


from skimage import exposure

# Adaptive Equalization
img_adapteq = exposure.equalize_adapthist(img, clip_limit=0.03)

plt.imshow(img_adapteq, cmap=plt.cm.gray)
plt.show()


(191, 384)
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 86
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 86
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 128
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 128
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 32
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 32
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 64
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 64
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 43
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 43
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 10
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 10
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 2
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 2
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 16
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 16
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 52
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 52
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 37
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 37
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 12
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 12
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 22
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 22
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 5
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 5
  n_excess -= hist[under].size
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:262: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 9
  hist[under] += 1
C:\Users\stree\Anaconda3\lib\site-packages\skimage\exposure\_adapthist.py:263: VisibleDeprecationWarning: boolean index did not match indexed array along dimension 0; dimension is 256 but corresponding boolean dimension is 9
  n_excess -= hist[under].size

In [89]:
patches = image.extract_patches_2d(img, (20, 20), max_patches=2, random_state=0)
patches.shape
plt.imshow(patches[0], cmap=plt.cm.gray)
plt.show()

from skimage.transform import rescale

im_small = rescale(img, 0.5)
patches = image.extract_patches_2d(im_small, (20, 20), max_patches=2, random_state=0)
patches.shape
plt.imshow(patches[0], cmap=plt.cm.gray)
plt.show()



In [92]:
count_vect = CountVectorizer(stop_words='english', ngram_range=(1,2))
X_train_counts = count_vect.fit_transform(twenty_train.data[0:1])
print(X_train_counts[0].toarray())
print(count_vect.vocabulary_.keys())


[[1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 3 1 1 1 1 1 1 2 1 1 1 1 2
  1 1 4 1 1 1 1 3 3 2 1 1 5 1 1 1 2 3 2 1 1 1 2 1 1 2 2 2 1 1 1 1 1 1 1 1 2
  1 1 1 1 1 1 1 1 1 1 1 1 3 3 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2
  1 1 1 1 2 2 1 1 3 3 1 1 1 1 1 1]]
dict_keys(['tuberlin zrz', '42 11', 'ftp 192', 'stefan', 'getting', 'binary prompt', 'board', 'ftp', 'mailgzrz', 'quit', 'wanna', '29 nntp', 'board cd', 'organization', 'software', '7900', 'ls binary', 'mailgzrz 1qpf1r', 'getting latest', 'mget', 'ftp cd', 'organization tuberlin', 'access', 'lines', '192', 'drivers genoa', 'drivers question', 'drivers ftp', 'ftp site', 'zrz lines', 'graphics', 'email best', 'graphics board', 'hartmann behse', 'question', 'best regards', 'berlin stefan', '192 109', 'host mikro', 'email', 'genoa', 'hi', 'behse subject', 'berlin', 'berlin hi', '9ti organization', '7900 board', 'latest software', 'tuberlin', 'latest', 'email harti', 'ee tu', 'ls', 'article', '109', 'prompt hash', 'hash', 'cards access', 'article mailgzrz', 'subject genoa', 'best', '109 42', 'ftp password', 'lines 29', 'zrz', 'cd 7000series', 'opened', 'password', 'nntp posting', '7000series', 'wanna latest', 'access ftp', 'regards stefan', 'host', 'site article', 'board drivers', 'quit sequence', 'subject', 'tu', 'ee', 'password ftp', 'tu berlin', 'latest drivers', 'opened ftp', 'posting', 'question email', 'drivers', '1qpf1r', 'sequence', 'hartmann', 'graphics cards', 'login ftp', 'harti mikro', 'mget quit', '42', 'behse', 'drivers 7900', 'genoa graphics', 'cd', 'hi opened', 'hash wanna', 'site getting', 'pub', '29', '1qpf1r 9ti', 'nntp', '7000series mget', 'cards', 'genoa ls', '9ti', 'hartmann email', '11 login', 'site', 'sequence drivers', 'mikro ee', '11', 'mikro', 'stefan hartmann', 'pub genoa', 'regards', 'login', 'software drivers', 'prompt', 'harti', 'posting host', 'binary', 'cd pub'])

In [ ]: