[PYTHON] Data analysis practice (data acquisition / organization / confirmation) -Dereste event 2001th border-
Overview
- I want to practice time series data analysis, so I will try it with "2001th borderline data of Dereste's event pt ranking".
- If possible, I would like to say, "Predict the 2001 line at the moment the event starts." (I have a harsh feeling)
- As a preliminary preparation, data acquisition, organization, and summary confirmation are performed on this page.
- The features are not enough, so I would like to add more and dig deeper.
- The scripts created this time are a1 ~ a4 of here.
Data used
[Imus Dereste Strategy Summary Wiki [Idolmaster Cinderella Girls Starlight Stage]](https://imascg-slstage-wiki.gamerch.com/%E3%82%A4%E3%83%99%E3%83%B3% E3% 83% 88% E3% 83% 87% E3% 83% BC% E3% 82% BF) data
Data summary
- Data items
| name | meaning | Variable scale | Data type |
|---|---|---|---|
| event name | Event name | Nominal scale | String |
| 2001 Border pt | 2001 Borderline points (I want to know about this) | Proportional scale | Numerical value (integer) |
| format | Event format (Attapon, Groove, Carnival) | Nominal scale | Character string (category) |
| attribute | Event attributes (format)=(Only for Groove) | Nominal scale | Character string (category) |
| date | Event start date | Interval scale | Date type |
| period | Event length (hours) | Proportional scale | Numerical value (integer) |
- Number of data: 95
- Missing
- Attribute: 63 missing. Missing except for format Groove
- Period: 3 missing. Missing if the format is Carnival
- Statistic
- Event name: No duplicate
- 2001 Border: No outliers. Mean is higher than median
- Format: 3 types. Up to 60 duplicates
- Date: No duplicates
- Attributes: 3 types. Up to 11 duplicates
- Period: No outliers. Mean is higher than median
| Event | 2001 | format | date | attribute | period | |
|---|---|---|---|---|---|---|
| count | 95 | 95 | 95 | 95 | 32 | 92 |
| unique | 95 | NaN | 3 | 95 | 3 | NaN |
| top | Passion fan fanfare | NaN | Attapon | 2015/12/4 | Vo | NaN |
| freq | 1 | NaN | 60 | 1 | 11 | NaN |
| mean | NaN | 95234.4 | NaN | NaN | NaN | 183.5 |
| std | NaN | 42973.7 | NaN | NaN | NaN | 19.1 |
| min | NaN | 40096 | NaN | NaN | NaN | 150 |
| 25% | NaN | 63761.5 | NaN | NaN | NaN | 174 |
| 50% | NaN | 83532 | NaN | NaN | NaN | 174 |
| 75% | NaN | 115178.5 | NaN | NaN | NaN | 198 |
| max | NaN | 224697 | NaN | NaN | NaN | 249 |
Data distribution
format
- Attapon format closes about two-thirds
- Carnival only 3 times
| format | The number of data |
|---|---|
| Carnival | 3 |
| Groove | 32 |
| Attapon | 60 |
attribute
- Data format is Groove only
- Held fairly evenly
| attribute | The number of data |
|---|---|
| Da | 10 |
| Vi | 11 |
| Vo | 11 |
2001 border
- All data
- Long hem to the right
- Color coded by format
- Carnival format is less frequent, but borders are higher
- The attack format has a long hem distribution to the right
- Groove has two mountains
period
- All data
- No data in Carnival format due to lack
Relationship between variables
By format
2001 border
- Carnival: Omitted because there are only 3 data
- Groove: Overall the lowest. Median <mean.
- Attapon: Middle in three formats. Median <mean. There are outliers on the top.
| format | Carnival | Groove | Attapon |
|---|---|---|---|
| count | 3 | 32 | 60 |
| mean | 188751.7 | 87048.6 | 94924.4 |
| std | 19012.5 | 35318.9 | 42349.2 |
| min | 176743 | 40096 | 42944 |
| 25% | 177791.5 | 52942.5 | 67515.8 |
| 50% | 178840 | 84560 | 80589.5 |
| 75% | 194756 | 114458 | 112983 |
| max | 210672 | 170014 | 224697 |
period
- Carnival: No data
- Groove: 174 for the lower half and above
- Attapon: min is smaller than Groove and max is larger than Groove. There are outliers on
| format | Carnival | Groove | Attapon |
|---|---|---|---|
| count | 0 | 32 | 60 |
| mean | NaN | 188.3 | 181.0 |
| std | NaN | 18.1 | 19.3 |
| min | NaN | 174 | 150 |
| 25% | NaN | 174 | 174 |
| 50% | NaN | 174 | 174 |
| 75% | NaN | 198 | 198 |
| max | NaN | 222 | 249 |
For each attribute
- Data format is Groove only
- 2001 rank border
- Da: Average, median, minimum, and maximum are all the lowest
- Vi: Average / minimum / maximum is the highest
- Vo: Highest median
| attribute | Da | Vi | Vo |
|---|---|---|---|
| count | 10 | 11 | 11 |
| mean | 81120.2 | 92813.3 | 86673.3 |
| std | 32024.0 | 38595.9 | 37182.3 |
| min | 40096 | 46300 | 42544 |
| 25% | 54767.3 | 64639.5 | 49871 |
| 50% | 78106.5 | 82143 | 100476 |
| 75% | 110899.3 | 110700 | 118864.5 |
| max | 127482 | 170014 | 140000 |
period
- Correlation matrix: seems to be uncorrelated
- I think that the borderline relationship is not zero because you can earn points if the period is long, but other factors seem to be stronger.
| 2001 | period | |
|---|---|---|
| 2001 | 1.0 | -0.126 |
| period | -0.126 | 1.0 |
date
2001 border
- All data
- Color coded by format
period
- Early events have exceeded 200h, but recently it has been fixed at 160h ~ 200h
Summary
- format
- If you want to dig deep, it seems better to start with the attacker with the largest amount of data.
- Carnival has a small amount of data, so analysis is difficult
- Period
- I thought that the longer the period, the higher the border, but it was not so
- I don't think the relationship is zero, but other factors may be stronger.
- Date
- As a tendency, it seems to be gradually rising
- The factors that come to mind are as follows
- The number of players has increased and the battle has become fierce
- It's easier to earn points
- The number of characters has increased (such as ability inflation and the number of people who can be left unattended)
- Grand live performance has improved time efficiency
- Other possible factors
- Variations for each event
- Popular event ranking reward idols