def recursive_tree_exploration(feature_set, past_accuracy, past_model):
last_added = feature_set[len(feature_set) - 1]
new_feature_set = copy.copy(feature_set)
#POR QUe AGREGAR 0 ??
## COUNT 1 APPEND
new_feature_set.append(0)
max_accuracy = past_accuracy
best_set = feature_set
best_model = past_model
for iter in xrange(last_added + 1, feature_amount):
## COUNT 1 POP
new_feature_set.pop()
## COUNT 1 APPEND
new_feature_set.append(iter)
new_x_norm, y, _, _ = getDataSubSet(new_feature_set)
model, calc_best_model, calc_accuracy = manual_cross_validation(
new_x_norm, y, models, names, True)
if calc_accuracy > max_accuracy:
max_accuracy = calc_accuracy
best_set = new_feature_set
best_model = calc_best_model
calc_accuracy, calc_best_set, calc_best_model = recursive_tree_exploration(
new_feature_set, max_accuracy, calc_best_model)
if calc_accuracy > max_accuracy:
max_accuracy = calc_accuracy
best_set = calc_best_set
best_model = calc_best_model
print best_set
print "BEST SET ACCURACY: " + str(max_accuracy)
print "BEST SET MODEL: " + str(best_model)
return max_accuracy, best_set, best_model
def tree_selection_heuristic():
# The feature set that is going to be evaluated
feature_set = [0]
max_accuracy = 0.0
best_set = []
best_model = None
# For every set of 1 feature the recursive search is applied
for i in xrange(1, feature_amount):
feature_set.pop()
feature_set.append(i)
new_x_norm, y, _, _ = getDataSubSet(feature_set)
#INIT FOR RECURSIVE CALL
model, calc_best_model, calc_accuracy = manual_cross_validation(
new_x_norm, y, models, names, True)
#RECURSIVE CALL
calc_accuracy, calc_best_set, calc_best_model = recursive_tree_exploration(
feature_set, calc_accuracy, calc_best_model)
if calc_accuracy > max_accuracy:
max_accuracy = calc_accuracy
best_set = calc_best_set
best_model = calc_best_model
print "-----------------------------------"
print "---------BEST FEATURE SET----------"
print best_set
print "-----------------------------------"
print "-------------ACCURACY--------------"
print max_accuracy
return best_model
size = pow(2,n)
# Vars to select the best suited model
bestModel = None
bestName = "none"
bestMean = 0.0
bestSet = []
for ii in xrange(size):
subset = []
for jj in xrange(n):
if (ii & (1 << jj)) > 0:
subset.append(set[jj])
print subset
if (len(subset)>0):
try:
X,y,X_test,Y_test= getDataSubSet(subset)
scaler = preprocessing.MinMaxScaler()
scaler.fit(X)
x_norm=scaler.transform(X)
#initialize models
forest= RandomForestClassifier(n_estimators=100, max_depth=20, random_state=111)
gdBoost = GradientBoostingClassifier(random_state=111)
mlp = MLPClassifier(solver='lbfgs', alpha=1e-5,hidden_layer_sizes=(10, 2), random_state=111)
models=[forest,gdBoost,mlp]
names= ["Random Forest", "Gradient Boosting", "MuliLayer Perceptrons"]
##Using all the features
import json
from sklearn import metrics as m
from sklearn import preprocessing
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.externals import joblib
from sklearn.feature_selection import RFECV
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import VarianceThreshold
from sklearn.feature_selection import chi2
from GetDataSet import getDataSubSet
from Validation import manual_cross_validation
X, y, X_test, Y_test = getDataSubSet(
[3, 5, 34, 35, 38, 39, 40, 42, 51, 52, 56, 60, 62, 99])
## Count number of 'easy' labeled instances and total instances
# This is done to keep control of the correct distribution of the dataset and the parameters of the experiment.
easyCount = 0
totalCount = 0
for i in xrange(len(Y_test)):
if Y_test[i] == "Easy":
easyCount += 1
totalCount += 1
print("Ratio of Easy over all on testing set: %0.2f" %
((easyCount + 0.0) / len(Y_test)))
easyCount = 0
for i in xrange(len(y)):
if y[i] == "Easy":
easyCount += 1
from imblearn.over_sampling import SMOTE from GetDataSet import getDataSubSet #set that defines the different sets of features that will be used #features = [1,2,3,4,5,6,7,44]+[x for x in xrange(8,15)]+[31,32,33,34,45,50]+[c for c in xrange(35,44)]+[56,57,58,60,65,67,68,84] #features =[3, 5, 34, 35, 38, 39, 40, 42, 51, 52, 56, 60, 62, 99] features = [x for x in range(100, 300)] print features # Vars to select the best suited model bestModel = None bestName = "none" bestMean = 0.0 bestSet = [] X, y, X_test, Y_test = getDataSubSet(features) scaler = preprocessing.MinMaxScaler() scaler.fit(X) x_norm = scaler.transform(X) #initialize models #forest= RandomForestClassifier(n_estimators=9100, max_depth=300, random_state=111) gdBoost = GradientBoostingClassifier(random_state=111) # mlp = MLPClassifier(solver='lbfgs', alpha=1e-5,hidden_layer_sizes=(10, 2), random_state=111) # models=[forest,gdBoost,mlp] # names= ["Random Forest", "Gradient Boosting", "MuliLayer Perceptrons"] # # # # ##Using all the features # model,name,mean=manual_cross_validation(x_norm, y, models, names, True)