Мой ИНС дает мне матрицу предсказаний вместо одного вектора предсказаний.предсказание «матрица» prob_pred выглядит следующим образом:
predict p0 p1
1 1 0.02001849 0.97998151
2 1 0.28398389 0.71601611
3 1 0.43267364 0.56732636
4 0 0.85280527 0.14719473
5 0 0.97794556 0.02205444
6 1 0.04342903 0.95657097
Правильно ли взять последний столбец матрицы (столбец 3) и сказать, что если> 0,5, он предсказал 1, а вместо этого его 0?Вот мой код:
# Splitting the dataset into the training set and test set
training_set = dataset[1:2740,]
test_set = dataset[2741:3748,]
# Feature Scaling - very important for ANN as it reduces the computational effort
training_set[-63] = scale(training_set[-63])
test_set[-63] = scale(test_set[-63])
# Fitting ANN to the Training set - h2o package is for classification (most efficient package)/neuralnet for regressors/nnet for models with only one layer
h2o.init(nthreads = -1) # minus one means i take all the available cores to use for the computation
classifier = h2o.deeplearning(y = 'Good.Bad.Stock',
training_frame = as.h2o(training_set),
activation = 'Rectifier',
hidden = c(30,30),
epochs = 100, # this is how many times the iteration should take place
train_samples_per_iteration = 1) # -2 here is auto-tuning the Gradint descent backpropogation (one could also have 0 for one epoch or -1 for batch) or simply put in 1 for the stochastic GD
# Predicting the Test set results
prob_pred = h2o.predict(classifier, newdata = as.h2o(test_set[-63])) # now we have probablities - but now we need to classify - so next line needed
prob_pred
y_pred = (prob_pred [,3] >= 0.5)
y_pred = as.vector(y_pred) # we need to make a standard vector - the h2o frame is not accepted in the CM
#maybe also apply summary function (if possible) to find out which independent variables are the most important ones
# Making the Confusion Matrix
cm_ann = table(test_set[, 63], y_pred)
cm_ann
Спасибо!