def getHeatmap(dataset_name):
#dataset_name = request.args.get("dataset_name")
print(dataset_name)
data = None
feature_names = None
if ("_updated" in dataset_name):
data = np.loadtxt(open("../data/" + str(dataset_name) + ".csv", "rb"),
delimiter=",",
skiprows=1)
feature_names = np.genfromtxt("../data/" + dataset_name +
".featurenames.csv",
delimiter=",",
dtype='str',
skip_header=1)
else:
data = np.loadtxt(open("./sample_data/" + str(dataset_name) + ".csv",
"rb"),
delimiter=",",
skiprows=1)
feature_names = np.genfromtxt("./sample_data/" + dataset_name +
".featurenames.csv",
delimiter=",",
dtype='str',
skip_header=1)
X = None
if ("_updated" in dataset_name):
X = data[:, 1:-3]
else:
X = data[:, :-3]
y = np.int_(data[:, -3])
unique_labels = np.unique(y)
print(X.shape)
_, n_feats = X.shape
n_labels = len(unique_labels)
first_cpc_mat = np.zeros((n_feats, n_labels))
feat_contrib_mat = np.zeros((n_feats, n_labels))
# 1. get the scaled feature contributions and first cPC for each label
ccpca = CCPCA(n_components=1)
for i, target_label in enumerate(unique_labels):
ccpca.fit(X[y == target_label],
X[y != target_label],
var_thres_ratio=0.2,
n_alphas=80,
max_log_alpha=0.2)
first_cpc_mat[:, i] = ccpca.get_first_component()
feat_contrib_mat[:, i] = ccpca.get_scaled_feat_contribs()
if (dataset_name == "mnist_updated"
or dataset_name == "fashion_mnist_updated"):
xlabel_names = [None] * n_labels
for i, label in enumerate(unique_labels):
if (unique_labels[label] == -1):
c = "Z"
else:
c = chr(65 + unique_labels[label])
xlabel_names[i] = str(c)
f = open(
"/home/user/Desktop/heatmap/data/" + dataset_name + "_labels.csv",
"w")
f.write("label\n")
for label in xlabel_names:
f.write(str(label) + "\n")
f.close()
print(xlabel_names)
ylabel_names = feature_names.tolist()
f = open(
"/home/user/Desktop/heatmap/data/" + dataset_name +
"_features.csv", "w")
f.write("feature\n")
for feature in ylabel_names:
f.write(str(feature) + "\n")
f.close()
f = open(
"/home/user/Desktop/heatmap/data/" + dataset_name + "_heatmap.csv",
"w")
f.write("feature,label,contribution\n")
for i, feature in enumerate(ylabel_names):
for j, label in enumerate(xlabel_names):
f.write(
str(feature) + "," + str(label) + "," +
str(feat_contrib_mat[i, j]) + "\n")
else:
# 2. apply optimal sign flipping
OptSignFlip().opt_sign_flip(first_cpc_mat, feat_contrib_mat)
# 3. apply hierarchical clustering with optimal-leaf-ordering
mr = MatReorder()
feat_contrib_mat = mr.fit_transform(feat_contrib_mat)
# 4. apply aggregation
n_feats_shown = n_feats
agg_feat_contrib_mat, label_to_rows, label_to_rep_row = mr.aggregate_rows(
feat_contrib_mat, n_feats_shown, agg_method='abs_max')
# plot cluster names
xlabel_names = [None] * n_labels
for i, label in enumerate(mr.order_col_):
if (unique_labels[label] == -1):
c = "Z"
else:
c = chr(65 + unique_labels[label])
xlabel_names[i] = str(c)
f = open("../data/" + dataset_name + "_labels.csv", "w")
f.write("label\n")
for label in xlabel_names:
f.write(str(label) + "\n")
f.close()
# plot feature names
ylabel_names = np.array(feature_names)[mr.order_row_]
# ylabel_names = np.array(feature_names, dtype=object)[label_to_rep_row]
# for i in range(len(ylabel_names)):
# name = ylabel_names[i]
# rows = label_to_rows[i]
# if len(rows) > 1:
# ylabel_names[i] = name + ', ' + str(len(rows) - 1) + ' more'
ylabel_names = ylabel_names.tolist()
f = open("../data/" + dataset_name + "_features.csv", "w")
f.write("feature\n")
for feature in ylabel_names:
f.write(str(feature) + "\n")
f.close()
f = open("../data/" + dataset_name + "_heatmap.csv", "w")
f.write("feature,label,contribution\n")
for i, feature in enumerate(ylabel_names):
for j, label in enumerate(xlabel_names):
f.write(
str(feature) + "," + str(label) + "," +
str(feat_contrib_mat[i, j]) + "\n")
f.close()
plt.figure()
colors = ['navy', 'turquoise', 'darkorange']
lw = 2
for color, i, target_name in zip(colors, [0, 1, 2], [0, 1, 2]):
plt.scatter(X_r[y == i, 0],
X_r[y == i, 1],
color=color,
alpha=.8,
lw=lw,
label=target_name)
plt.legend(loc='best', shadow=False, scatterpoints=1)
plt.title('cPCA of IRIS dataset (alpha=2.15)')
plt.show()
ccpca = CCPCA()
ccpca.fit(X[y == 0], X[y != 0], var_thres_ratio=0.5, max_log_alpha=0.5)
X_r2 = ccpca.transform(X)
plt.figure()
for color, i, target_name in zip(colors, [0, 1, 2], [0, 1, 2]):
plt.scatter(X_r2[y == i, 0],
X_r2[y == i, 1],
color=color,
alpha=.8,
lw=lw,
label=target_name)
plt.legend(loc='best', shadow=False, scatterpoints=1)
plt.title('ccPCA of IRIS dataset (alpha =' + str(ccpca.get_best_alpha()) + ')')
plt.show()
def wordCloudGen():
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from ccpca import CCPCA
from opt_sign_flip import OptSignFlip
from mat_reorder import MatReorder
# print("I was here")
# classLabel = request.get.params("label")
dataset_name = request.args.get("datasetName")
print(dataset_name)
data = None
feature_names = None
if ("_updated" in dataset_name):
data = np.loadtxt(open("../data/" + str(dataset_name) + ".csv", "rb"),
delimiter=",",
skiprows=1)
feature_names = np.genfromtxt("../data/" + dataset_name +
".featurenames.csv",
delimiter=",",
dtype='str',
skip_header=1)
else:
data = np.loadtxt(open("./sample_data/" + str(dataset_name) + ".csv",
"rb"),
delimiter=",",
skiprows=1)
feature_names = np.genfromtxt("./sample_data/" + dataset_name +
".featurenames.csv",
delimiter=",",
dtype='str',
skip_header=1)
print(feature_names)
X = None
if ("_updated" in dataset_name):
X = data[:, 1:-3]
else:
X = data[:, :-3]
y = np.int_(data[:, -3])
unique_labels = np.unique(y)
target_label = 0
ccpca = CCPCA(n_components=2)
ccpca.fit(X[y == target_label],
X[y != target_label],
var_thres_ratio=0.5,
n_alphas=40,
max_log_alpha=0.5)
# get results
cpca_result = ccpca.transform(X)
best_alpha = ccpca.get_best_alpha()
cpca_fcs = ccpca.get_feat_contribs()
X = data[:, :-3]
y = np.int_(data[:, -3])
unique_labels = np.unique(y)
_, n_feats = X.shape
n_labels = len(unique_labels)
first_cpc_mat = np.zeros((n_feats, n_labels))
feat_contrib_mat = np.zeros((n_feats, n_labels))
ccpca = CCPCA(n_components=1)
for i, target_label in enumerate(unique_labels):
ccpca.fit(X[y == target_label],
X[y != target_label],
var_thres_ratio=0.5,
n_alphas=40,
max_log_alpha=0.5)
first_cpc_mat[:, i] = ccpca.get_first_component()
feat_contrib_mat[:, i] = ccpca.get_scaled_feat_contribs()
OptSignFlip().opt_sign_flip(first_cpc_mat, feat_contrib_mat)
mr = MatReorder()
mr.fit_transform(feat_contrib_mat)
print(feature_names)
combined = np.vstack((feature_names, cpca_fcs)).T
print(combined)
pd.DataFrame(combined).to_csv("../data/featContrib.csv")
resp = make_response()
resp.headers['Access-Control-Allow-Origin'] = '*'
return resp