def test_single_variate_single_dimension(self):
# this is a single-variable expressed as a 1-d numpy array
# (each element of the array is an instance)
X = np.array([1, 2, 3, 4, 5])
B = 2
s = np.mean
bootstrap = Bootstrap.Bootstrap(X, s, B)
bootstrap.run()
assert bootstrap.N == 5
assert bootstrap.B == B
def test_07a(self):
X = my_data.get_data()
#print(X)
s = self.ratio_first_eigenvector_to_sum
B = 200
#B = 10
# explore the empirical data...
covariance_matrix = np.cov(X, bias=True, rowvar=False)
w, v = LA.eig(covariance_matrix)
v = np.transpose(v)
print("--- empirical data - shape ---")
print(X.shape)
print("--- empirical data - covariance matrix ---")
print(covariance_matrix)
print("--- empirical data - eigenvalues ---")
print(w)
print("--- empirical data - eigenvectors ---")
print(v)
# prepare to collect data - empty 3-d array
num_attributes = X.shape[1]
self.eigenvectors = np.empty([0, num_attributes, num_attributes])
# run the bootstrap
print("--- run the bootstrap ---")
bootstrap = Bootstrap.Bootstrap(X, s, B)
bootstrap.add_callback(self.my_callback)
[std, sem] = bootstrap.run()
print("standard deviation:")
print(std)
print("standard error of the mean")
print(sem)
#assert(False)
# investigate the results
# plot the theta_stars (the measure) from the bootstrap replications
# the expectation is this is somewhat gaussian (long tails are not acceptable)
print("--- results from bootstrap ---")
my_charts.plot_histogram(bootstrap.theta_star, "Count of Occurrences",
"Ratio: eigenV1/sum(eigen)",
"Histogram - Count of EigenV1/sum")
#assert(False)
# plot the first two principal component vectors using box-and-whisker
# we are looking for (lack of) variability
print("first two principal components")
print(self.eigenvectors.shape)
my_charts.plot_box_and_whisker()
def test_treatment(self):
treatment = np.array([94, 197, 16, 38, 99, 141, 23])
s = self.my_s
B = 100
# look at the original data...
print("Treatment sample size: ", treatment.shape[0], " mean: ", np.mean(treatment), "sem: ", stats.sem(treatment))
# run the bootstrap
### this is incorrect - we actually should run boot strap
# on the DIFFERENCE btween Treatment and control
bootstrap = Bootstrap.Bootstrap(treatment, s, B)
[std, sem] = bootstrap.run()
def test_single_variate(self):
# this is a single-variable expressed as a 2-d numpy array
# (the attribute is column 0, rows are instances)
num_instances = 100
num_attributes = 1
X = np.random.randint(5, size=(num_instances, num_attributes))
s = np.mean
B = 2
bootstrap = Bootstrap.Bootstrap(X, s, B)
bootstrap.run()
assert bootstrap.N == num_instances
assert bootstrap.B == B
def test_multi_variate(self):
# this is a three-variable expressed as a 2-d numpy array
# (the attributes are columns, rows are instances)
num_instances = 6
num_attributes = 3
X = np.random.randint(5, size=(num_instances, num_attributes))
s = np.mean
B = 2
bootstrap = Bootstrap.Bootstrap(X, s, B)
bootstrap.run()
assert bootstrap.N == num_instances
assert bootstrap.B == B
def generate_bootstraps(self): # bootstrapping
for n in range(0, self.nTree):
b = Bootstrap()
b.generate(self.original_dataset)
self.bootstraps.append(b)
return self.bootstraps
BOOTSTRAP_N = 20 # number of bootstrap samples (YOU CAN PLAY AROUND WITH THIS)
DATA_START_INDEX = 1 # account for df's named index column 0 (DON'T CHANGE THIS UNLESS YOUR DATASET NEEDS IT)
DO_K_SWEEP = True # switch to do sweep of K values using K means to find optimal K
OPTIMAL_K = 3 # Iris dataset has 3 clusters (ground truth), change this for different datasets
# import data
iris = datasets.load_iris()
df = pd.DataFrame(data=np.c_[iris['data']], columns=iris['feature_names'])
# prepare data (add index column 'flower')
prep = Prepare('flower', len(df)).names_join(df)
df = prep['df']
labels = prep['labels']
# generate bootstrap samples
bts = Bootstrap(df, BOOTSTRAP_SIZE, BOOTSTRAP_N).get_bootstraps()
# determine optimal clustering K
kmeans = Bootstrap.kmeans_bootstrap(bts, DO_K_SWEEP, BOOTSTRAP_N,
DATA_START_INDEX, OPTIMAL_K, MAX_K)
# max k determined above becomes optimal k
gmm = RunAlgos(3, BOOTSTRAP_N, DATA_START_INDEX, bts, kmeans).run_GMM()
agglomerative = RunAlgos(3, BOOTSTRAP_N, DATA_START_INDEX, bts,
kmeans).run_Agglomerative()
kmeans_ = RunAlgos(3, BOOTSTRAP_N, DATA_START_INDEX, bts, kmeans).run_KMeans()
# consensus clustering
cc_init = Consensus(kmeans_, gmm, agglomerative, bts, df, DATA_START_INDEX, 3,
labels)
mats = cc_init.combine_results()
def main(args):
import Bootstrap as app
global app
app = app.Bootstrap(args)
app.exec_()
