import numpy as np
import sklearn as sklearn
import pdb as check
import modelfunctions
#set name of weights/biases file
weightFile = '../Data/TheBestWeights.h5'
predFile = './Predicted.h5'
testFile = '../Data/Testing.h5'
#names of features
field_names_in = [
'stresses_full_xx', 'stresses_full_yy', 'stresses_full_xy',
'stresses_full_xz', 'stresses_full_yz', 'stresses_full_zz'
]
#name of label
field_names_out = 'aftershocksyn'
#load model
model = modelfunctions.create_model()
#assess performance of training network on testing data set
model.load_weights(weightFile)
inTrue, outTrue = modelfunctions.LoadInputs(testFile, field_names_in,
field_names_out)
outPred = model.predict(inTrue)
modelfunctions.writeHDF(predFile, inTrue, outPred)
auc = sklearn.metrics.roc_auc_score(outTrue, outPred)
print 'merged AUC on testing data set: ' + str(auc)
from keras.callbacks import ModelCheckpoint import matplotlib.pyplot as plt #set name of weights file and training/testing files weightFile = '../Data/TheBestWeights.h5' trainFile = '../Data/Training.h5' testFile = '../Data/Testing.h5' #names of features field_names_in = ['stresses_full_xx', 'stresses_full_yy', 'stresses_full_xy', 'stresses_full_xz','stresses_full_yz','stresses_full_zz'] #name of label field_names_out = 'aftershocksyn' #load training data set IN_Train, OUT_Train = modelfunctions.LoadInputs(trainFile, field_names_in, field_names_out) #pull out some validation data with positive grid cells upsampled so that validation data has equal numbers of positive and negative samples posidx = np.where(OUT_Train==1) numpos = np.size(posidx) negidx = np.where(OUT_Train==0) numneg = np.size(negidx) #divide data into positive and negative samples POSdata = np.column_stack((IN_Train[posidx,:][0], OUT_Train[posidx].T)) NEGdata = np.column_stack((IN_Train[negidx,:][0], OUT_Train[negidx].T)) np.random.seed(42) #shuffle order of samples (same way every time to ensure validation data does not change) np.random.shuffle(POSdata) np.random.shuffle(NEGdata) np.random.seed() #reseed