vectorization-maps/fmnist.py at master · Cimagroup/vectorization-maps · GitHub

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
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
from sklearn.model_selection import GridSearchCV
import vectorization as vect
from auxiliary_functions import *
from numpy.random import seed
import pickle

s=1
seed(s)

normalization=False
n_iters = 100


from fashion_mnist import mnist_reader
path_feat = "fashion_mnist/features/"
path_diag= "fashion_mnist/pdiagrams/"
path_results = "results/"


vec_parameters = dict()
vec_parameters['GetPersStats']=(),
vec_parameters['GetAlgebraicFunctions']=(),
vec_parameters['GetEntropySummary'] = [[15,30,50]]
vec_parameters['GetBettiCurveFeature'] = [[15,30,50]]
vec_parameters['GetPersLifespanFeature'] = [[15,30,50]]
vec_parameters['GetAtolFeature'] = [[2,4,8,16]]
vec_parameters['GetPersImageFeature'] = [[0.05,0.5,1],[3,6,12,20]]
vec_parameters['GetPersSilhouetteFeature'] = [[15,30,50], [0,1,2,5]]
vec_parameters['GetComplexPolynomialFeature'] = [[3, 5, 10],['R', 'S', 'T']]
vec_parameters['GetPersLandscapeFeature'] = [[15,30,50], [1,2,3,5]]
vec_parameters['GetTemplateFunctionFeature'] = [[2,3,5,10], [.5, 1, 2]]
vec_parameters['GetPersTropicalCoordinatesFeature'] = [[10,50,250]]
vec_parameters['GetAdaptativeSystemFeature'] = [['gmm'],
                                                [3,4,5,10,15]]


onlyForest = [
    {'base_estimator': ['RF'], 'n_estimators': [50,100]},
 ]

searchG = GridSearchCV(
    main_classifier(), param_grid=onlyForest, cv=5,
    return_train_score=True, scoring='accuracy'
)


_, y_train = mnist_reader.load_mnist('fashion_mnist/data/fashion', kind='train')
_, y_test = mnist_reader.load_mnist('fashion_mnist/data/fashion', kind='t10k')

train_index = list(range(len(y_train)))
test_index = list(range(len(y_train), len(y_train)+len(y_test)))
index = train_index+test_index


pdiagrams = dict()

for i in index:
    #pdiagrams["pdiag_taxi_l_"+str(i)]= safe_load(path_diag + "taxi_l_"+str(i))
    pdiagrams["pdiag_taxi_u_"+str(i)]= safe_load(path_diag + "taxi_u_"+str(i))


func_list = [getattr(vect, keys) for keys in vec_parameters.keys()]
for func in func_list:
    feature_dictionary = dict()
    vec_methods = dict()
    vec_methods[func.__name__] = vec_parameters[func.__name__]

    feature_dictionary=feature_computation(vec_methods,pdiagrams,"pdiag_taxi_u_",
                                           train_index, test_index)

    with open(path_results+'FMNIST_feature_'+func.__name__+'.pkl', 'wb') as f:
        pickle.dump(feature_dictionary, f)

    best_scores=parameter_optimization(train_index, y_train, vec_methods,
                                       feature_dictionary, searchG, normalization)

    print("Parameter optimization:",best_scores)

    with open(path_results+'FMNIST_best_scores_'+func.__name__+'.pkl', 'wb') as f:
      pickle.dump(best_scores, f)

    train_scores, test_scores = scores(train_index, y_train, test_index, y_test,
                                       vec_methods, feature_dictionary, best_scores,
                                       n_iters, normalization)

    print("The train accuracy is", train_scores)
    print("The test accuracy is", test_scores)

    with open(path_results+'FMNIST_train_scores_'+func.__name__+'.pkl', 'wb') as f:
      pickle.dump(train_scores, f)
    with open(path_results+'FMNIST_test_scores_'+func.__name__+'.pkl', 'wb') as f:
      pickle.dump(test_scores, f)