使用sci-kit Learn从ANN检索矩阵

Question

USing python 3.5, I do the following

from sklearn.neural_network import MLPRegressor
import pandas as pd

from sklearn import datasets from sklearn.decomposition import PCA

# import some data to play with
iris = datasets.load_iris()
X = iris.data[0:100, :2]  # we only take the first two features.
Y = iris.target[0:100]

X = StandardScaler().fit_transform(X)

X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.15)
reg = MLPRegressor(hidden_layer_sizes=(10,),
                           solver="lbfgs", #solver="lbfgs",
                           max_iter = 10000,
                           activation = 'relu')
reg.fit(X_train,y_train)

i can get the matrix using reg.coefs_[0] but when I multiply it by a test case, I don't get 0 or 1

what I'm trying to do is train an ANN based on lab data, and then implement it as matrix multiplication on a piece of hardware. The hardware can do basic arithmetic, so I need to deconstruct my ANN into a matrix, and then program matrix multiplication as iterated multiplication/addition

Answer 1

The short, "specific to this question" answer is

np.dot((np.dot(X_test[0],reg.coefs_[0]) +reg.intercepts_[0] ),reg.coefs_[1]) + reg.intercepts_[1]

The long answer is as follows (even longer version here: https://www.mohammadathar.com/blog/2017/2/15/a-different-look-at-neural-networks )

A perceptron neural net can be modeled like this:

What that really means, is "multiply input values by a weight, and add. then add in a bias value. Then activate. Then multiply by weights, and add. Then activate. And repeat!"

As matrix math, it looks like this

So the reg.coefs_ is the matrix coefficients (per layer) and the reg.intercepts_ is the activation values (again, per layer)