# -*- coding: utf-8 -*-
"""
Created on Mon Jul 14 16:11:37 2014
@author: Christoph Lassner
"""
import os
import sys
sys.path.insert(0, os.path.join('..', '..', 'build', 'bindings', 'python'))
import numpy as np
import fertilized
soil = fertilized.Soil('d', 'd', 'd', fertilized.Result_Types.regression)
tree = soil.StandardRegressionTree(1, 0, 1, 5, 2, 4)
X = np.array([[0.], [0.]])
Y = np.array([[0.], [10**3]])
tree.fit(X, Y)
print('Tree: %f, STD: %f' %
(np.sqrt(tree.predict(np.array([[0.]]))[0, 1]), np.std(Y[:, 0])))
if os.name == 'nt':
tree.save('tree_windows.ft')
else:
tree.save('tree_linux.ft')
import numpy as np
np.random.seed(8)
from plottools import make_spiral, point_prob_plot, probaproxy, accuracy_score
#plt.rc('text', usetex=True)
#plt.rc('font', family='serif')
np.random.seed(1)
#%% Prepare...
n_classes = 2
n_trees = 200
ploty = [-6, 6, 100]
plotx = [-6, 6, 100]
X, Y = make_spiral(n_arms=n_classes, noise=.4)
# Forest.
soil = f.Soil()
dec_name = 'quadratic'
rs = 1
suggestions = 5
depth = 5
if dec_name == 'aligned':
feat_sel_prov = soil.StandardFeatureSelectionProvider(2,
1,
2,
2,
random_seed=rs)
else:
feat_sel_prov = soil.StandardFeatureSelectionProvider(1, 2, 2, 2)
if dec_name == 'aligned':
feat_calc = soil.AlignedSurfaceCalculator()
elif dec_name == 'linear':
if INTERACTIVE:
import matplotlib.pyplot as plt
import numpy as np
from plottools import make_spiral, point_prob_plot, accuracy_score
np.random.seed(1)
# Prepare...
n_classes = 3
n_trees = 200
ploty = [-6, 6, 100]
plotx = [-6, 6, 100]
X, Y = make_spiral(n_arms=n_classes, noise=.4)
#%%
soil = f.Soil('float', 'float', 'uint', f.Result_Types.probabilities)
for depth in [3, 6, 15]:
cls = []
lm = []
for i in range(n_trees):
cls.append(soil.ThresholdDecider(
soil.StandardFeatureSelectionProvider(1, 2, 2, 2),
# soil.AlignedSurfaceCalculator(),
# soil.LinearSurfaceCalculator(400, random_seed=1+i),
soil.QuadraticSurfaceCalculator(400,
np.array([X[:, 0].min(),
X[:, 0].max(),
X[:, 1].min(),
X[:, 1].max()], dtype='float'),
random_seed=1+i),
soil.ClassificationThresholdOptimizer(True,
os.path.join(root_dir, image_dir,
'image-%03d.png' % (neg_idx)))))
assert not images[-1] is None
annotations[pos_offset + neg_number * n_samples_neg:pos_offset +
(neg_number + 1) * n_samples_neg,
0] = neg_number + len(train_pos_ids)
annotations[pos_offset+neg_number*n_samples_neg : pos_offset+(neg_number+1)*n_samples_neg, 1] = \
np.random.randint(patch_size[1] // 2,
images[-1].shape[1] - patch_size[1] // 2,
size=(n_samples_neg,))
annotations[pos_offset+neg_number*n_samples_neg : pos_offset+(neg_number+1)*n_samples_neg, 2] = \
np.random.randint(patch_size[0] // 2,
images[-1].shape[0] - patch_size[0] // 2,
size=(n_samples_neg,))
#%% Feature extraction.
soil = fertilized.Soil('uint8', 'int16', 'int16',
fertilized.Result_Types.hough_map)
if os.path.isfile('feat_images.pkl'):
with open('feat_images.pkl', 'r') as f:
feat_images = pickle.load(f)
else:
cvimages = []
for im in images:
if im.ndim == 2:
cvimages.append(np.ascontiguousarray(skimage.color.gray2rgb(im)))
else:
cvimages.append(np.ascontiguousarray(im[:, :, :3]))
feat_images = [
soil.extract_hough_forest_features(im, full=(n_feature_channels == 32))
for im in cvimages
]
with open('feat_images.pkl', 'w') as f:
"""
Created on Fri Mar 13 11:40:48 2015
@author: christoph
"""
import os
import sys
sys.path.insert(0, os.path.join('..', '..', 'build', 'bindings', 'python'))
import numpy as np
import numpy.testing as npt
import fertilized
###############################################################################
# Check the availability (since it's an optional module).
soil = fertilized.Soil()
try:
extractor = soil.DNNFeatureExtractor(True) # CPU only.
except:
print("The library has been built without caffe. Features cannot be extracted.")
sys.exit(0)
###############################################################################
# Prepare the images.
from PIL import Image
timage_zeros = np.zeros((227, 227, 3), dtype='float32')
timage_ones = np.ones((227, 227, 3), dtype='float32')
timage_opencv = np.array(Image.open(os.path.join(os.path.dirname(__file__), 'sample-resized.png'))).astype('float32')
timage_opencv = np.ascontiguousarray(timage_opencv[:, :, ::-1])
# Assume the image has had its means subtracted already.
size=(n_samples, input_dim)).astype('float32')
samplesY = (samplesX**2) - 0.2 * samplesX - 1
noise_var = 2.3
random_noise = np.random.randn(n_samples, annot_dim) * noise_var
samplesY += random_noise
# Define the plotted space
minX = samplesX.min()
maxX = samplesX.max()
deltaX = (maxX - minX)
minY = samplesY.min()
maxY = samplesY.max()
deltaY = (maxY - minY)
plotX = [minX - 0.2 * deltaX, maxX + 0.2 * deltaX, 100]
plotY = [minY - 0.2 * deltaY, maxY + 0.2 * deltaY, 100]
# Create and fit a <float> regression forest
soil = f.Soil('f', 'f', 'f', f.Result_Types.regression)
n_trees = 10
max_depth = 2
n_thresholds = 2
forest = soil.StandardRegressionForest(input_dim, max_depth, 1, n_thresholds,
n_trees, 3, 6)
forest.fit(samplesX, samplesY, 8)
# Plot the result
plt.figure()
forest_probabilistic_plot(forest, samplesX, samplesY, plotX, plotY)
plt.savefig('regression_prob.png')
plt.show()