119 lines
3 KiB
Python
119 lines
3 KiB
Python
|
#!/usr/bin/env python3
|
||
|
|
# -*- coding: utf-8 -*-
|
||
|
|
|
||
|
|
import matplotlib.pyplot as plt
|
||
|
|
from sklearn.neural_network import MLPClassifier
|
||
|
|
from sklearn.model_selection import train_test_split
|
||
|
|
from sklearn.model_selection import KFold
|
||
|
|
from sklearn.metrics import precision_score
|
||
|
|
import random
|
||
|
|
import time
|
||
|
|
|
||
|
|
"""
|
||
|
|
from sklearn.datasets import fetch_openml
|
||
|
|
mnist = fetch_openml('mnist_784')
|
||
|
|
"""
|
||
|
|
|
||
|
|
"""
|
||
|
|
|
||
|
|
images = mnist.data.values.reshape((-1, 28, 28))
|
||
|
|
plt.imshow(images[0],cmap=plt.cm.gray_r,interpolation="nearest")
|
||
|
|
plt.show()
|
||
|
|
|
||
|
|
print(mnist.target[0])
|
||
|
|
"""
|
||
|
|
|
||
|
|
|
||
|
|
indices = [i for i in range(len(mnist.data))]
|
||
|
|
random.shuffle(indices)
|
||
|
|
indices = indices[:15000]
|
||
|
|
|
||
|
|
data = [mnist.data.values[i] for i in indices]
|
||
|
|
target = [mnist.target[i] for i in indices]
|
||
|
|
|
||
|
|
bestClf = None
|
||
|
|
bestScore = 0
|
||
|
|
scores = []
|
||
|
|
|
||
|
|
train_sizes = [0.05*i for i in range(1,20)]
|
||
|
|
|
||
|
|
|
||
|
|
xtrain, xtest, ytrain, ytest = train_test_split(data, target, train_size=0.7)
|
||
|
|
|
||
|
|
"""
|
||
|
|
network = MLPClassifier(hidden_layer_sizes = (50))
|
||
|
|
network.fit(xtrain, ytrain)
|
||
|
|
ytest_pred = clf.predict(xtest)
|
||
|
|
print("Average = macro:")
|
||
|
|
print(precision_score(ytest, ytest_pred, average = "macro"))
|
||
|
|
print("Average = None:")
|
||
|
|
print(precision_score(ytest, ytest_pred, average = None))
|
||
|
|
|
||
|
|
print(target[4])
|
||
|
|
print(network.predict(data[4].reshape(1,-1)))
|
||
|
|
plt.imshow(data[4].reshape(28, 28), cmap=plt.cm.gray_r, interpolation="nearest")
|
||
|
|
plt.show()
|
||
|
|
"""
|
||
|
|
|
||
|
|
"""
|
||
|
|
n_layers = [i for i in range(1,51)]
|
||
|
|
|
||
|
|
|
||
|
|
for n in n_layers:
|
||
|
|
print(n)
|
||
|
|
network = MLPClassifier(hidden_layer_sizes = (50,)*n)
|
||
|
|
network.fit(xtrain, ytrain)
|
||
|
|
scores += [network.score(xtest, ytest)]
|
||
|
|
|
||
|
|
plt.plot(n_layers, scores)
|
||
|
|
"""
|
||
|
|
|
||
|
|
"""
|
||
|
|
for _ in range(5):
|
||
|
|
n_layers = random.randint(1,3)
|
||
|
|
layer_size = tuple([random.randint(10,300) for _ in range(n_layers)])
|
||
|
|
print(layer_size)
|
||
|
|
|
||
|
|
network = MLPClassifier(hidden_layer_sizes = layer_size, solver = "adam", activation = "relu")
|
||
|
|
network.fit(xtrain, ytrain)
|
||
|
|
print(f"training time = {time.time() - start_time}, score = {network.score(xtest, ytest)}")
|
||
|
|
"""
|
||
|
|
|
||
|
|
"""
|
||
|
|
solvers = ["lbfgs", "sgd", "adam"]
|
||
|
|
for s in solvers:
|
||
|
|
print(s + ":")
|
||
|
|
start_time = time.time()
|
||
|
|
network = MLPClassifier(hidden_layer_sizes = (50,)*5, solver = s)
|
||
|
|
network.fit(xtrain, ytrain)
|
||
|
|
print(f"training time = {time.time() - start_time}, score = {network.score(xtest, ytest)}")
|
||
|
|
"""
|
||
|
|
|
||
|
|
"""
|
||
|
|
functions = ["identity", "logistic", "tanh", "relu"]
|
||
|
|
|
||
|
|
for f in functions:
|
||
|
|
print(f + ":")
|
||
|
|
start_time = time.time()
|
||
|
|
network = MLPClassifier(hidden_layer_sizes = (50,)*5, activation = f)
|
||
|
|
network.fit(xtrain, ytrain)
|
||
|
|
print(f"training time = {time.time() - start_time}, score = {network.score(xtest, ytest)}")
|
||
|
|
"""
|
||
|
|
|
||
|
|
times = []
|
||
|
|
scores = []
|
||
|
|
|
||
|
|
for i in range(-7,7):
|
||
|
|
alpha = 10**i
|
||
|
|
print(f"alpha = {alpha}:")
|
||
|
|
start_time = time.time()
|
||
|
|
network = MLPClassifier(hidden_layer_sizes = (50,)*5, alpha = alpha)
|
||
|
|
network.fit(xtrain, ytrain)
|
||
|
|
trainingTime = time.time() - start_time
|
||
|
|
score = network.score(xtest, ytest)
|
||
|
|
times +=[trainingTime]
|
||
|
|
scores += [score]
|
||
|
|
|
||
|
|
plt.plot([i for i in range(-7,7)], times)
|
||
|
|
plt.show()
|
||
|
|
plt.plot([i for i in range(-7,7)], scores)
|