Perceptron/test.py at master · DiodDan/Perceptron · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
from perceptron import *
import numpy as np
from progress.bar import IncrementalBar

export_file = "nums_recognition_MNIST4"

f = open("Mnist_databse.txt", "r")
json_d = json.loads(f.read())
f.close()
print("file read")
inputs = [(np.array(json_d["data"][i]["image"]) - 127, [1 if json_d["data"][i]["answer"] == j else 0 for j in range(10)]) for i in range(70000)]
split_point = 60000
learn_inputs = inputs[:split_point]
batch_learn_inputs = [learn_inputs[i * 1000:(i+1) * 1000] for i in range(60)]
test_inputs = inputs[split_point:]
print("Inputs ready")
network = NeuralNetwork(learn_speed=0.8, sensors_amount=784, hidden_layer_amount=2, neuron_amount_in_hidden_layers=[16] * 2, result_neurons_amount=10)
print(f"Network ready(after learning weights can de seen on {export_file})")
r_ans = 0
for era in range(100):
    for batch in batch_learn_inputs:
        learn_progress = IncrementalBar(f"Learn progress for Era(batch):{era}", max=1000)
        for learn_input in batch:
            network.set_input(input_data=list(learn_input[0]), answer=learn_input[1])
            network.forward_propagation(clear_error=False)
            network.batch_delta_weights_calculation()
            learn_progress.next()
        network.batch_weights_adjustment()
        learn_progress.finish()
    r_ans = 0
    test_progress = IncrementalBar(f"Test progress for Era:{era}", max=10000)
    for test in test_inputs[:10000]:
        network.set_input(input_data=list(test[0]), answer=test[1])
        network.forward_propagation()
        arr = [network.result_neurons_arr[i].value for i in range(len(network.result_neurons_arr))]
        arr1 = network.answer
        if arr.index(max(arr)) == arr1.index(max(arr1)):
            r_ans += 1
        test_progress.next()
    test_progress.finish()
    if r_ans / 100 >= 95:
        network.export_weights()
    print(f"Era:{era} answered right on {r_ans / 100}%", "\n")


for era in range(100, 130):
    learn_progress = IncrementalBar(f"Learn progress for Era(batch full input):{era}", max=len(learn_inputs))
    for learn_input in learn_inputs:
        network.set_input(input_data=list(learn_input[0]), answer=learn_input[1])
        network.forward_propagation(clear_error=False)
        network.batch_delta_weights_calculation()
        learn_progress.next()
    network.batch_weights_adjustment()
    learn_progress.finish()
    r_ans = 0
    test_progress = IncrementalBar(f"Test progress for Era:{era}", max=10000)
    for test in test_inputs[:10000]:
        network.set_input(input_data=list(test[0]), answer=test[1])
        network.forward_propagation()
        arr = [network.result_neurons_arr[i].value for i in range(len(network.result_neurons_arr))]
        arr1 = network.answer
        if arr.index(max(arr)) == arr1.index(max(arr1)):
            r_ans += 1
        test_progress.next()
    test_progress.finish()
    print(f"Era:{era} answered right on {r_ans / 100}%", "\n")

for test in test_inputs[:1000]:
    network.set_input(input_data=list(test[0]), answer=test[1])
    network.forward_propagation()
    arr = [network.result_neurons_arr[i].value for i in range(len(network.result_neurons_arr))]
    print(f"network is {max(arr) * 100 // 1} % sure:", arr.index(max(arr)))
    arr = network.answer
    print("answer:", arr.index(max(arr)))
network.export_weights(export_file)