[PYTHON] Promised victory ("Hobby x Machine Learning")

Introduction

This is the first post. I will write down the material that I have prepared for LT in the company. (I will do my best to continue to output ...)

table of contents

1. Preface / Purpose
2. System flow
3. Details
4. Announcement of practical results
5. Conclusion
6. TODO

1. 1. Preface / Purpose

Title: ** "Hobby x Machine Learning" **

It's a big title, My hobby is ** Pachislot ** ..

What is pachislot?

It is a slot machine installed in pachinko parlors, and is named "rotary type game machine" by law. It is called a pachislot because it is a slot machine in a pachinko parlor. There is a set of ** 6 stages (*) ** in pachislot, which is manually changed by a pachinko clerk after the store is closed.

• Not all models have 6 levels.

I will explain with a concrete example. "Go Go Juggler KK (Kita Denshi)" (I think many of you know this face.)

The basic specifications of this stand are as follows.

As I explained earlier, there is a setting on the slot base, It is set to approximately ** 6 levels **. And as the setting goes up, the specifications of the machine discount stand also go up.

This setting is ** manually changed by a pachinko parlor clerk every night **.

• It may be deferred.

So I thought.

If you set it manually, Isn't ** law ** born unconsciously ...?

So, the purpose

Collect pachislot data, use machine learning to guide the rules, and be sure to win.

2. System-wide flow

To explain this system insanely about, the composition is as follows. I will explain each step by step.

3. 3. Details

① Data collection

This system is implemented using Python.

environment Windows10 + Anaconda3

** Library used **

-selenium 3.141.0 (crawling) -Beautifulsoup 4.7.1 (scraping)

** Collection period ** 2019/7/15 - 2019/9/14

• Collected every day at 23:20

** Data actually collected **

■ Stumbling point

The site where you can see the pachinko ball output information was only accessible from your smartphone.

Specify the device in the chromedriver option ** Resolve **

mobile_emulation = { "deviceName": "Nexus 5" }
chrome_options = webdriver.ChromeOptions()
chrome_options.add_experimental_option("mobileEmulation", mobile_emulation)
driver = webdriver.Chrome("C:/chromedriver.exe", desired_capabilities = chrome_options.to_capabilities())
driver.get("URL of a pachinko parlor")

② Machine learning

Use scikit-learn's "Random Forest".

This time, I stopped the data and learned by expanding the juggler. Predict how many settings the platform will have on that day from the trained model! !!

-** Objective variable : Setting - Explanatory variables **: Date, day of the week

Preprocessing that is not smart at all

#After messing around with the messy dataframe,

#Guess the setting from the probability (it is called the guess value even though it is the objective variable)...）
tmp = line[1][8].split("/")
hoge = float(tmp[0]) / float(tmp[1])
if hoge >= my_pro[5]:
   hoge_np = np.append(hoge_np, 5)
elif hoge >= my_pro[4]:
   hoge_np = np.append(hoge_np, 4)
elif hoge >= my_pro[3]:
   hoge_np = np.append(hoge_np, 3)
elif hoge >= my_pro[2]:
   hoge_np = np.append(hoge_np, 2)
elif hoge >= my_pro[1]:
   hoge_np = np.append(hoge_np, 1)
else:
   hoge_np = np.append(hoge_np, 0)

#There are many other mess...

If you mess with it, throw it into the learning model

#Separate for training and evaluation
x_train, x_test, y_train, y_test = model_selection.train_test_split(x, y, test_size=0.2)

#Learning
model = RandomForestClassifier()
model.fit(x_train, y_train)

③ Practice

■ Practice day 2019/10/5 (Sat)

■ Location A certain pachinko parlor

You can definitely win by hitting the 1257 juggler (setting is 5 !!)! !!

So

Four. Practical result announcement

When I usually go to hit, I take about three people, Sakichi. I can win this time, so I went to practice with Satoshi! !!

The result is ...! !! !!

Yes, I was able to bring two Sakichi brilliantly.

Five. Conclusion

Not so sweet.

6．TODO Based on the following points of reflection, we will brush up so that we can definitely win.

Aim for a millionaire! !!

• If you lose next time, you will retire.

Reflections

-The explanatory variables are not plausible. -The settings of the previous day and the settings of the left and right stands should be considered. -You should observe the store trends. -The number of data is not large in the first place. -I don't know if there is a tendency. -Even if you can guess the settings, you cannot always win. ** (Darkness of Pachislot) **