Related
png files are usually index values assosiated with a default palette. by default, index values can be read by PIL image, for example:
import tensorflow as tf
from PIL import Image
import numpy as np
path = '1.png'
image1 = Image.open(path)
print(image1.mode)
array1 = np.array(image1)
print(array1.shape)
print(set(list(array1.reshape(-1))))
the results:
P
(1024, 543)
{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11}
index can be converted to palette colors:
image2 = image1.convert('RGB')
print(image2.mode)
array2 = np.array(image2)
print(array2.shape)
print(set([tuple(x) for x in list(array2.reshape(-1, 3))]))
the results are
RGB
(1024, 543, 3)
{(0, 255, 255), (255, 255, 0), (0, 0, 255), (85, 255, 170), (170, 255, 85), (255, 0, 0), (255, 170, 0), (0, 170, 255), (0, 85, 255), (255, 85, 0), (0, 0, 170)}
the question is, by default, tensorflow 2.x API read palette colors, is there a way to read the index values?
tensor = tf.io.read_file(path)
tensor = tf.image.decode_png(tensor, channels=3)
array3 = tensor.numpy()
print(array3.shape)
print(set([tuple(x) for x in list(array3.reshape(-1, 3))]))
(1024, 543, 3)
{(0, 255, 255), (255, 255, 0), (0, 0, 255), (85, 255, 170), (170, 255, 85), (255, 0, 0), (255, 170, 0), (0, 170, 255), (0, 85, 255), (255, 85, 0), (0, 0, 170)}
my temporary solution is to remove all the palette assosiated with png files. Still looking for more efficient solutions
from PIL import Image
import numpy as np
img = Image.open('1.png')
img = np.array(img)
img = Image.fromarray(img)
img.save('2.png')
Ideal solution
A clean solution would be to re-implement a custom op to decode a PNG without palette conversion.
Currently, the palette conversion in TF for the decode_png-op is done at core level:
// convert palette to rgb(a) if needs be.
if (context->color_type == PNG_COLOR_TYPE_PALETTE)
png_set_palette_to_rgb(context->png_ptr);
Work-around solution
(TF2.X)
As you've mentioned PIL in your sample code, you could wrap the PIL-call with tf.py_function in order to get the desired behavior, like:
def read_png(mask):
def read_fn(p):
return np.asarray(Image.open(p.numpy().decode()))
return tf.py_function(read_fn, [mask], tf.uint8)
and then read a paletted PNG-image without automatic palette application, like:
img = read_png(path)
img= img.numpy()
print(img.shape)
print(set(list(img.reshape(-1))))
Sample output
(1024, 543)
{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11}
Note: in TF1.X this works only in graph-mode, and you would have to use tf.py_func to wrap the TF around PIL-function (+ some minor changes).
I use a histogram to display the distribution. Everything works fine if the spacing of the bins is uniform. But if the interval is different, then the bar width is appropriate (as expected). Is there a way to set the width of the bar independent of the size of the bins ?
This is what i have
This what i trying to draw
from matplotlib import pyplot as plt
my_bins = [10, 20, 30, 40, 50, 120]
my_data = [5, 5, 6, 8, 9, 15, 25, 27, 33, 45, 46, 48, 49, 111, 113]
fig1 = plt.figure()
ax1 = fig1.add_subplot(121)
ax1.set_xticks(my_bins)
ax1.hist(my_data, my_bins, histtype='bar', rwidth=0.9,)
fig1.show()
I cannot mark your question as a duplicate, but I think my answer to this question might be what you are looking for?
I'm not sure how you'll make sense of the result, but you can use numpy.histogram to calculate the height of your bars, then plot those directly against an arbitrary x-scale.
x = np.random.normal(loc=50, scale=200, size=(2000,))
bins = [0,1,10,20,30,40,50,75,100]
fig = plt.figure()
ax = fig.add_subplot(211)
ax.hist(x, bins=bins, edgecolor='k')
ax = fig.add_subplot(212)
h,e = np.histogram(x, bins=bins)
ax.bar(range(len(bins)-1),h, width=1, edgecolor='k')
EDIT Here's with the adjustment to the x-tick labels so that the correspondence is easier to see.
my_bins = [10, 20, 30, 40, 50, 120]
my_data = [5, 5, 6, 8, 9, 15, 25, 27, 33, 45, 46, 48, 49, 111, 113]
fig = plt.figure()
ax = fig.add_subplot(211)
ax.hist(my_data, bins=my_bins, edgecolor='k')
ax = fig.add_subplot(212)
h,e = np.histogram(my_data, bins=my_bins)
ax.bar(range(len(my_bins)-1),h, width=1, edgecolor='k')
ax.set_xticks(range(len(my_bins)-1))
ax.set_xticklabels(my_bins[:-1])
I have a tensor of allowed items
index = tf.constant([61, 215, 23, 18, 241, 125])
and need to remove items from input sequence batches that are not in index.
seq = tf.constant(
[
[ 18, 241, 0, 0],
[125, 61, 23, 241],
[ 23, 92, 18, 0],
[ 5, 61, 215, 18],
]
)
After the calculation in this case I need
result_needed = tf.constant(
[
[ 18, 241, 0, 0],
[125, 61, 23, 241],
[ 23, 18, 0, 0],
[ 61, 215, 18, 0],
]
)
I cannot do this in Python because this calculation happens during predictions. Also note that while item IDs here are small, solution needs to deal with numbers from 1 to 2^40.
Answer
After some serious pondering time, I came up with the following:
idx_range = tf.reshape(tf.range(seq.shape[-2]), [-1, 1])
idx_tile = tf.tile(idx_range, [1, seq.shape[-2].value])
idx_flat = tf.reshape(idx_tile, [-1])
truth_value = tf.equal(index, tf.expand_dims(seq, -1))
one_hot = tf.to_float(truth_value)
ones = tf.nn.top_k(tf.reduce_sum(one_hot, -1), seq.shape[-1]).indices
ones_flat = tf.reshape(ones, [-1])
ones_idx = tf.reshape(
tf.stack([idx_flat, ones_flat], axis=1),
tf.concat([seq.shape, [2]], axis=0)
)
tf.gather_nd(seq, ones_idx)
This is not exactly what I said I needed, but actually got me close enough. Instead of the output replacing the blacklisted items with 0, it moves them to the end. If you needed them gone, I'm sure there's a method to remove them, but I'm not looking into it. Apologies.
I'm trying to get to grips with sci-kit learn for some simple machine learning projects but I'm coming unstuck with Pipelines and wonder what I've done wrong...
I'm trying to work through a tutorial on Kaggle
Here's my code:
import pandas as pd
train = pd.read_csv(local path to training data)
train_labels = pd.read_csv(local path to labels)
from sklearn.decomposition import PCA
from sklearn.svm import LinearSVC
from sklearn.grid_search import GridSearchCV
pca = PCA()
clf = LinearSVC()
n_components = arange(1, 39)
loss =['l1','l2']
penalty =['l1','l2']
C = arange(0, 1, .1)
whiten = [True, False]
from sklearn.pipeline import Pipeline
#set up pipeline
pipe = Pipeline(steps=[('pca', pca), ('clf', clf)])
#set up GridsearchCV
estimator = GridSearchCV(pipe, dict(pca__n_components = n_components, pca__whiten = whiten,
clf__loss = loss, clf__penalty = penalty, clf__C = C))
estimator
Returns:
GridSearchCV(cv=None,
estimator=Pipeline(steps=[('pca', PCA(copy=True, n_components=None, whiten=False)), ('clf', LinearSVC(C=1.0, class_weight=None, dual=True, fit_intercept=True,
intercept_scaling=1, loss='l2', multi_class='ovr', penalty='l2',
random_state=None, tol=0.0001, verbose=0))]),
fit_params={}, iid=True, loss_func=None, n_jobs=1,
param_grid={'clf__penalty': ['l1', 'l2'], 'clf__loss': ['l1', 'l2'], 'clf__C': array([ 0. , 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]), 'pca__n_components': array([ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,
35, 36, 37, 38]), 'pca__whiten': [True, False]},
pre_dispatch='2*n_jobs', refit=True, score_func=None, scoring=None,
verbose=0)
But when I try to train data:
estimator.fit(train, train_labels)
The error is:
428 for test_fold_idx, per_label_splits in enumerate(zip(*per_label_cvs)):
429 for label, (_, test_split) in zip(unique_labels, per_label_splits):
--> 430 label_test_folds = test_folds[y == label]
431 # the test split can be too big because we used
432 # KFold(max(c, self.n_folds), self.n_folds) instead of
IndexError: too many indices for array
Can anyone point me in the right direction?
It turns out that the Pandas dataframe is the wrong shape.
estimator.fit(train.values, train_labels[0].values)
works, although I also had to drop the penalty term.
i am newbie and just want to implement Hierarchical Agglomerative clustering for RGB images. For this I extract all values of RGB from an image. And I process image.Next I find its distance and then develop the linkage. Now from linkage I want to extract my original data (i.e RGB values) on specified indices with indices id. Here is code I have done so far.
image = Image.open('image.jpg')
image = image.convert('RGB')
im = np.array(image).reshape((-1,3))
rgb = list(im.getdata())
X = pdist(im)
Y = linkage(X)
I = inconsistent(Y)
based on the 4th column of consistency. I opt minimum value of the cutoff in order to get maximum clusters.
cutoff = 0.7
cluster_assignments = fclusterdata(Y, cutoff)
# Print the indices of the data points in each cluster.
num_clusters = cluster_assignments.max()
print "%d clusters" % num_clusters
indices = cluster_indices(cluster_assignments)
ind = np.array(enumerate(rgb))
for k, ind in enumerate(indices):
print "cluster", k + 1, "is", ind
dendrogram(Y)
I got results like this
cluster 6 is [ 6 11]
cluster 7 is [ 9 12]
cluster 8 is [15]
Means cluster 6 contains the indices of 6 and 11 leafs. Now at this point I stuck in how to map these indices to get original data(i.e rgb values). indices of each rgb values to each pixel in the image. And then I have to generate codebook to implement Agglomeration Clustering. I have no idea how to approach this task. Read a lot of stuff but nothing clued.
Here is my solution:
import numpy as np
from scipy.cluster import hierarchy
im = np.array([[54,101,9],[ 67,89,27],[ 67,85,25],[ 55,106,1],[ 52,108,0],
[ 55,78,24],[ 19,57,8],[ 19,46,0],[ 95,110,15],[112,159,57],
[ 67,118,26],[ 76,127,35],[ 74,128,30],[ 25,62,0],[100,120,9],
[127,145,61],[ 48,112,25],[198,25,21],[203,11,10],[127,171,60],
[124,173,45],[120,133,19],[109,137,18],[ 60,85,0],[ 37,0,0],
[187,47,20],[127,170,52],[ 30,56,0]])
groups = hierarchy.fclusterdata(im, 0.7)
idx_sorted = np.argsort(groups)
group_sorted = groups[idx_sorted]
im_sorted = im[idx_sorted]
split_idx = np.where(np.diff(group_sorted) != 0)[0] + 1
np.split(im_sorted, split_idx)
output:
[array([[203, 11, 10],
[198, 25, 21]]),
array([[187, 47, 20]]),
array([[127, 171, 60],
[127, 170, 52]]),
array([[124, 173, 45]]),
array([[112, 159, 57]]),
array([[127, 145, 61]]),
array([[25, 62, 0],
[30, 56, 0]]),
array([[19, 57, 8]]),
array([[19, 46, 0]]),
array([[109, 137, 18],
[120, 133, 19]]),
array([[100, 120, 9],
[ 95, 110, 15]]),
array([[67, 89, 27],
[67, 85, 25]]),
array([[55, 78, 24]]),
array([[ 52, 108, 0],
[ 55, 106, 1]]),
array([[ 54, 101, 9]]),
array([[60, 85, 0]]),
array([[ 74, 128, 30],
[ 76, 127, 35]]),
array([[ 67, 118, 26]]),
array([[ 48, 112, 25]]),
array([[37, 0, 0]])]