notebook.community

Edit and run 





In [1]:



    


import sys
sys.path.append('../src')
import settings
sys.path.append(settings.CAFFE_PYTHON_PATH)
import caffe
import dataset
import cub_utils
import utils
import rects
import parts
import numpy as np
import matplotlib.pylab as plt
%matplotlib inline

import sklearn.svm
import sklearn.metrics
import sklearn.grid_search
from datetime import datetime as dt

from storage import datastore
from deep_extractor import CNN_Features_CAFFE_REFERENCE, Berkeley_Extractor



    











In [2]:



    


cub = dataset.CUB_200_2011(settings.CUB_ROOT)
all_img_infos = cub.get_all_image_infos()
all_seg_infos = cub.get_all_segmentation_infos()
IDtrain, IDtest = cub.get_train_test_id()
cub_parts = cub.get_parts()
cub_bbox = cub.get_bbox()

Baseline





In [3]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_ccrftt_1-60000')))
Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(feature_extractor.extract_one)



    











In [4]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_cccftt_1-60000')))
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(feature_extractor.extract_one)



    











In [9]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('ccr')))
Xtrain_r, ytrain, Xtest_r, ytest = cub.get_train_test(feature_extractor.extract_one)



    











In [10]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('ccc')))
Xtrain_c, ytrain, Xtest_c, ytest = cub.get_train_test(feature_extractor.extract_one)



    











In [5]:



    


Xtrain_h, ytrain, Xtest_h, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccphrf_def_unif_1-60000'))).extract_one)



    











In [6]:



    


Xtrain_b, ytrain, Xtest_b, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccpbrf_def_unif_1-60000'))).extract_one)



    











In [7]:



    


Xtrain_c_ft_rf, ytrain, Xtest_c_ft_rf, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccrf_def_1-60000'))).extract_one)

R





In [15]:



    


Xtrain_final = Xtrain_r
Xtest_final = Xtest_r

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.485329651363 0.489643

R-ft





In [16]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_ccrftt_1-60000')))
Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(feature_extractor.extract_one)

Xtrain_final = Xtrain_r_ft
Xtest_final = Xtest_r_ft

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.52002071108 0.524735

















In [17]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_ccrftt_2-60000')))
Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(feature_extractor.extract_one)

Xtrain_final = Xtrain_r_ft
Xtest_final = Xtest_r_ft

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.530721435968 0.53576

















In [18]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_ccrftt_3-60000')))
Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(feature_extractor.extract_one)

Xtrain_final = Xtrain_r_ft
Xtest_final = Xtest_r_ft

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.529168104936 0.533501

C





In [36]:



    


Xtrain_final = Xtrain_c
Xtest_final = Xtest_c

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.57352433552 0.578022

C-ft





In [19]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_cccftt_1-60000')))
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(feature_extractor.extract_one)

Xtrain_final = Xtrain_c_ft
Xtest_final = Xtest_c_ft

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.641525716258 0.644514

















In [20]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_cccftt_2-60000')))
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(feature_extractor.extract_one)

Xtrain_final = Xtrain_c_ft
Xtest_final = Xtest_c_ft

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.640490162237 0.643284

















In [21]:



    


feature_extractor = CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_cccftt_3-60000')))
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(feature_extractor.extract_one)

Xtrain_final = Xtrain_c_ft
Xtest_final = Xtest_c_ft

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.638591646531 0.641345

R+C





In [24]:



    


Xtrain_final = np.hstack((Xtrain_r, Xtrain_c))
Xtest_final = np.hstack((Xtest_r, Xtest_c))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.599240593718 0.603098

R-ft+C-ft





In [3]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_ccrftt_1-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_cccftt_1-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.660165688643 0.663286

















In [4]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_ccrftt_2-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_cccftt_2-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.655850880221 0.658868

















In [5]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_ccrftt_3-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(datastore(settings.storage('f_cccftt_3-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.660856057991 0.663518

DeepRF





In [9]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_1-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccftt_1-60000'))).extract_one)
Xtrain_h, ytrain, Xtest_h, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccphrf_def_unif_1-60000'))).extract_one)
Xtrain_b, ytrain, Xtest_b, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccpbrf_def_unif_1-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_h, Xtrain_b))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft, Xtest_h, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.735933724543 0.738757

















In [10]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_2-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccftt_2-60000'))).extract_one)
Xtrain_h, ytrain, Xtest_h, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccphrf_def_unif_2-60000'))).extract_one)
Xtrain_b, ytrain, Xtest_b, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccpbrf_def_unif_2-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_h, Xtrain_b))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft, Xtest_h, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.731791508457 0.734157

















In [11]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_3-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccftt_3-60000'))).extract_one)
Xtrain_h, ytrain, Xtest_h, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccphrf_def_unif_3-60000'))).extract_one)
Xtrain_b, ytrain, Xtest_b, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccpbrf_def_unif_3-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_h, Xtrain_b))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft, Xtest_h, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.737832240249 0.740365

SVM parameter





In [31]:



    


Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_h, Xtrain_b))
model = sklearn.grid_search.GridSearchCV(sklearn.svm.LinearSVC(), {'C': [1e0, 1e-1, 1e-2, 1e-3, 1e-4, 1e-5, 1e-6]}, n_jobs=3, cv=5)
model.fit(Xtrain_final, ytrain)
print model.best_params_



    


















    






{'C': 0.01}

















In [32]:



    


model.grid_scores_



    


















    
Out[32]:








[mean: 0.95963, std: 0.00441, params: {'C': 1.0},
 mean: 0.95963, std: 0.00441, params: {'C': 0.1},
 mean: 0.95979, std: 0.00434, params: {'C': 0.01},
 mean: 0.95963, std: 0.00398, params: {'C': 0.001},
 mean: 0.95796, std: 0.00294, params: {'C': 0.0001},
 mean: 0.94011, std: 0.00435, params: {'C': 1e-05},
 mean: 0.87154, std: 0.00354, params: {'C': 1e-06}]

Effect of C





In [41]:



    


Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft))

for c in [1e1, 1e0, 1e-1, 1e-2, 1e-3, 1e-4, 1e-5, 1e-6]:
    model = sklearn.svm.LinearSVC(C=c)
    model.fit(Xtrain_final, ytrain)
    preds = model.predict(Xtest_final)
    print c, sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






10.0 0.650155333103 0.653106
1.0 0.650155333103 0.653106
0.1 0.650155333103 0.653106
0.01 0.650155333103 0.653183
0.001 0.654124956852 0.657028
0.0001 0.660165688643 0.663286
1e-05 0.645150155333 0.648536
1e-06 0.589920607525 0.593018

Oracle Parts

Oh-ft (oracle head)

The oracle head method is better than all the baseline methods, only using head





In [9]:



    


Xtrain_oh, ytrain, Xtest_oh, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('ccpheadft-100000'))).extract_one)

Xtrain_final = Xtrain_oh
Xtest_final = Xtest_oh

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.667932343804 0.671085

Ob-ft (oracle body)





In [10]:



    


Xtrain_ob, ytrain, Xtest_ob, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('ccpbodyft-100000'))).extract_one)

Xtrain_final = Xtrain_ob
Xtest_final = Xtest_ob

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.527959958578 0.528481

R_ft + C_ft + Oh_ft





In [6]:



    


Xtrain_oh, ytrain, Xtest_oh, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('ccpheadft-100000'))).extract_one)
Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_1-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccftt_1-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_oh))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft, Xtest_oh))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.782706247843 0.784847

















In [7]:



    


Xtrain_oh, ytrain, Xtest_oh, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('ccpheadft-100000'))).extract_one)
Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_2-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccftt_2-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_oh))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft, Xtest_oh))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.784086986538 0.786089

















In [8]:



    


Xtrain_oh, ytrain, Xtest_oh, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('ccpheadft-100000'))).extract_one)
Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_3-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccftt_3-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_oh))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft, Xtest_oh))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.783914394201 0.785887

Finally Final

Complete(R_ft, C_RF_FT, P_H_RF_FT, P_B_RF_FT)

experiment 1





In [12]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_1-60000'))).extract_one)
Xtrain_c_ft_rf, ytrain, Xtest_c_ft_rf, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccrf_def_1-60000'))).extract_one)
Xtrain_h, ytrain, Xtest_h, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccphrf_def_unif_1-60000'))).extract_one)
Xtrain_b, ytrain, Xtest_b, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccpbrf_def_unif_1-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft_rf, Xtrain_h, Xtrain_b))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft_rf, Xtest_h, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.721263375906 0.723472

















In [14]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_2-60000'))).extract_one)
Xtrain_c_ft_rf, ytrain, Xtest_c_ft_rf, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccrf_def_1-60000'))).extract_one)
Xtrain_h, ytrain, Xtest_h, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccphrf_def_unif_2-60000'))).extract_one)
Xtrain_b, ytrain, Xtest_b, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccpbrf_def_unif_2-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft_rf, Xtrain_h, Xtrain_b))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft_rf, Xtest_h, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.717811529168 0.720146

















In [15]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_3-60000'))).extract_one)
Xtrain_c_ft_rf, ytrain, Xtest_c_ft_rf, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccrf_def_1-60000'))).extract_one)
Xtrain_h, ytrain, Xtest_h, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccphrf_def_unif_3-60000'))).extract_one)
Xtrain_b, ytrain, Xtest_b, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccpbrf_def_unif_3-60000'))).extract_one)

Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft_rf, Xtrain_h, Xtrain_b))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft_rf, Xtest_h, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.714532274767 0.716918

Which Part is the most important (in DeepRF)





In [16]:



    


Xtrain_r_ft, ytrain, Xtest_r_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccrftt_1-60000'))).extract_one)
Xtrain_c_ft, ytrain, Xtest_c_ft, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_cccftt_1-60000'))).extract_one)
Xtrain_h, ytrain, Xtest_h, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccphrf_def_unif_1-60000'))).extract_one)
Xtrain_b, ytrain, Xtest_b, ytest = cub.get_train_test(CNN_Features_CAFFE_REFERENCE(
    datastore(settings.storage('f_ccpbrf_def_unif_1-60000'))).extract_one)

- R





In [19]:



    


Xtrain_final = np.hstack((Xtrain_c_ft, Xtrain_h, Xtrain_b))
Xtest_final = np.hstack((Xtest_c_ft, Xtest_h, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.731273731446 0.733825

- C





In [20]:



    


Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_h, Xtrain_b))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_h, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.711080428029 0.71366

- Head





In [21]:



    


Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_b))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft, Xtest_b))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.676907145323 0.679486

- Body





In [22]:



    


Xtrain_final = np.hstack((Xtrain_r_ft, Xtrain_c_ft, Xtrain_h))
Xtest_final = np.hstack((Xtest_r_ft, Xtest_c_ft, Xtest_h))

model = sklearn.svm.LinearSVC(C=1e-4)
model.fit(Xtrain_final, ytrain)
preds = model.predict(Xtest_final)

print sklearn.metrics.accuracy_score(ytest, preds), utils.mean_accuracy(ytest, preds)



    


















    






0.730410769762 0.733428

















In [ ]:

Content source: yassersouri/omgh

Similar notebooks:

notebook.community | gallery | about