def get_prim(n_clusters, f_value, data, xstring, ystring, inputs, data_norm):
df = clustering(n_clusters, data, xstring, ystring, data_norm)
classes = df[['class']].drop_duplicates()['class']
list_prim = []
sorted(classes)
for i in range(1, max(classes) + 1):
list_prim.append(
prim.Prim(inputs, (df['class'] == i),
threshold=0.5,
threshold_type=">"))
boxes = []
for i in range(0, len(list_prim)):
boxes.append(list_prim[i].find_box())
obj = (f_value * boxes[i].peeling_trajectory['coverage'] -
(1 - f_value) * boxes[i].peeling_trajectory['density'])**2
if 1 in boxes[i].peeling_trajectory['coverage']:
coverage1 = np.where(
boxes[i].peeling_trajectory['coverage'] == 1)[0][0]
obj = obj.drop(obj.index[[coverage1]])
k = obj.argmin()
boxes[i].select(k)
return [boxes, df]
def compute_box_memberships(self, kde_noise_variance=.0001):
self.catted_x_tr = np.concatenate(self.x_tr)
kde_data = self.catted_x_tr + np.random.multivariate_normal(np.zeros(self.catted_x_tr.shape[1]), kde_noise_variance * np.eye(2)) #prevent singular matrix apparently some datapoints are identical/overplotted
kde_data = kde_data.reshape([kde_data.shape[1], kde_data.shape[0]])
kde = gaussian_kde(kde_data)
grid_for_prim = self.get_grid_for_prim()
density_estimate = kde(grid_for_prim.reshape([grid_for_prim.shape[1], grid_for_prim.shape[0]]))
max_density = np.amax(density_estimate)
print(density_estimate[0:20])
primmer = prim.Prim(
pd.DataFrame(grid_for_prim),
density_estimate,
threshold=max_density * self.gate_init_params['prim_threshold_percent']
)
#for i in range(self.gate_init_params['n_boxes']):
#primmer.find_box()
primmer.find_all()
print(primmer.limits, 'bark')
box_memberships_tr = self.get_box_memberships_tr(primmer._boxes)
self.box_memberships_tr = box_memberships_tr
import numpy as np
import matplotlib.pyplot as plt
import re
from os import rename
import glob as gb
# set solution number
solution_num = '1713'
utility = 'Cary'
metric = 'C_all'
# load objectives and establish regional criteria
perturbed = pd.read_csv('../Data/satisficing_' + str(solution_num) +
'_prim.csv')
DVs = pd.read_csv('../Data/all_perturbations' + solution_num + '.csv')
if utility == "Durham":
util_DVs = DVs[['D_RT', 'D_TT', 'D_Cont', 'D_Ins', 'D_JLA']]
elif utility == "Raleigh":
util_DVs = DVs[['R_RT', 'R_TT', 'R_Cont', 'R_Ins', 'R_JLA']]
else:
util_DVs = DVs[['C_RT', 'C_Cont', 'C_Ins', 'C_JLA']]
p = prim.Prim(util_DVs, perturbed[metric], threshold=819, threshold_type=">")
box = p.find_box()
print box
box.show_tradeoff()
plt.show()
plt.show()
#%% scenario discovery - PRIM
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from ema_workbench.analysis import prim
from ema_workbench import ema_logging
ema_logging.log_to_stderr(ema_logging.INFO)
x = experiments1.iloc[:, 0:50]
outcomesdf = pd.DataFrame.from_dict(outcomes1)
y = outcomesdf['Expected Number of Deaths'] < 1.0
prim_alg = prim.Prim(x, y, threshold=0.842, peel_alpha=0.1)
box1 = prim_alg.find_box()
box1.show_tradeoff()
plt.show()
prim_alg = prim.Prim(x, y, threshold=0.842, peel_alpha=0.1)
box2 = prim_alg.find_box()
box2.show_pairs_scatter(10)
plt.show()
plt.savefig('prim expected deaths<1.png')
#%% feature scoring
from ema_workbench.analysis import feature_scoring
from __future__ import print_function
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import prim
df = pd.DataFrame(np.random.rand(1000, 3), columns=["x1", "x2", "x3"])
p = prim.Prim(df,
lambda x : x["x1"]*x["x2"] + 0.3*x["x3"],
threshold=0.5,
threshold_type=">")
box = p.find_box()
box.show_tradeoff()
plt.show()
sns.distplot(outputs.AAPE) sns.pairplot(outputs) outputs['AAD_bin'] = np.where(outputs['AAD'] < 3700000, 1, 0) #Assing binary sum(outputs['AAD_bin']) outputs['AAPE_bin'] = np.where(outputs['AAPE'] < 94, 1, 0) #Assing binary sum(outputs['AAPE_bin']) # Look at smaller number of pass/fails - i.e. only small number are under threshold, so interrogate this # If '<' used, y values less than thresh will be 1's (cases of interest), all values graeter 0's ##https://github.com/Project-Platypus/PRIM/blob/master/README.md # AAPE OBJECTIVE # In this instance we are interested in cases where the response is greater than 0.5 # (as indicated by the "threshold" and "threshold_type" arguments) p = prim.Prim(inputs2, outputs.AAPE_bin, threshold=0.5, threshold_type=">") box = p.find_box() box.show_tradeoff() plt.show() box.select(42) print(box) box.show_details() # AAD OBJECTIVE p = prim.Prim(inputs1, outputs.AAD_bin, threshold=0.5, threshold_type=">") box = p.find_box() box.show_tradeoff() plt.show() box.select(42)
resp = robjects.FloatVector(response) h=r.hist(resp) hf = robjects.conversion.converter.rpy2py(h) pf = pd.DataFrame(hf) print(h) print(pf) print(pf.attrs) #print(prim_response) #thr = robjects.FloatVector([1.0,2.0]) #rprim = r['prim.box'] #prim_res = rprim(x=qf,y=prim_response,threshold=thr) #print(prim_res) p = prim.Prim(df, response, threshold=1.0, threshold_type=">") #box = p.find_box() #box.show_tradeoff() #plt.show() from sklearn.datasets import load_iris from sklearn import tree X, y = load_iris(return_X_y=True) print(X) print(y) clf = tree.DecisionTreeClassifier() clf = clf.fit(X, y) tree.plot_tree(clf)
from __future__ import print_function
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import prim
# Import dataset
df_all = pd.read_csv("data_comp_exp_ML.csv")
# Check for missing values
df_all.isnull().sum()
region = "asia"
df = df_all.query("region =='{}'".format(region))
# Declare feature vector and target variable
X = df.drop(columns=["resultados_2_c","resultados_3_c","region","policy","parameter_set","gcm","key","goal_2_c","goal_3_c"])
response = df["goal_2_c"]
p = prim.Prim(X,
response,
threshold=1,
threshold_type="=")
box = p.find_box()
box.show_tradeoff()
plt.show()
matriz[i][j] = coluna
matriz[j][i] = coluna
j = j + 1
i = i + 1
k = 0
for linha in matriz:
vetor = {}
i = 0
for j in range(len(linha)):
if linha[i] != 0:
vetor[str(j)] = linha[i]
i = i + 1
lista_encadeada[str(k)] = vetor
k = k + 1
for i, k in lista_encadeada.items():
print i + "-" + str(k)
#print matriz
grafo = grafos.Grafo(lista_encadeada)
#kruskal = kruskal.Kruskal(grafo)
#valor=0
#print kruskal.arvore_minima()
#for i in kruskal.arvore_minima():
# valor = valor + int(i[0])
#print valor
prim = prim.Prim(grafo)
print prim.arvore_minima('0')