板情報の可視化（フル板）
目次
板情報の可視化（フル板）¶
注意
ToDo: 解説を書く
import numpy as np
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
Tardis.dev よりフル板のCSVをダウンロードして読み込みます。
df = pd.read_csv("coinbase_incremental_book_L2_2019-07-01_BTC-USD.csv.gz", nrows=100000)
df.to_pickle("coinbase_incremental_book_L2_2019-07-01_BTC-USD.pickle", protocol=4)
注意
実際には上記のコードで取得しますが、ここでは取得済みのpickleファイルから読み込みます。
df = pd.read_pickle("coinbase_incremental_book_L2_2019-07-01_BTC-USD.pickle")
df.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 100000 entries, 0 to 99999
Data columns (total 8 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 exchange 100000 non-null object
1 symbol 100000 non-null object
2 timestamp 100000 non-null int64
3 local_timestamp 100000 non-null int64
4 is_snapshot 100000 non-null bool
5 side 100000 non-null object
6 price 100000 non-null float64
7 amount 100000 non-null float64
dtypes: bool(1), float64(2), int64(2), object(3)
memory usage: 5.4+ MB
df.loc[:, "timestamp"] = pd.to_datetime(df.loc[:, "timestamp"], unit="us")
df.loc[:, "local_timestamp"] = pd.to_datetime(df.loc[:, "local_timestamp"], unit="us")
df.sort_values("timestamp", inplace=True)
df.set_index("timestamp", drop=False, inplace=True)
df.index.name = None
# amount=0を欠損値とする
df.loc[df.loc[:, "amount"] == 0, "amount"] = np.nan
df.head()
| exchange | symbol | timestamp | local_timestamp | is_snapshot | side | price | amount | |
|---|---|---|---|---|---|---|---|---|
| 2019-07-01 00:00:00.923 | coinbase | BTC-USD | 2019-07-01 00:00:00.923 | 2019-07-01 00:00:01.444699 | False | bid | 10751.56 | 5.345 |
| 2019-07-01 00:00:00.924 | coinbase | BTC-USD | 2019-07-01 00:00:00.924 | 2019-07-01 00:00:01.444749 | False | bid | 10751.56 | 1.000 |
| 2019-07-01 00:00:00.926 | coinbase | BTC-USD | 2019-07-01 00:00:00.926 | 2019-07-01 00:00:01.444836 | False | ask | 10775.89 | NaN |
| 2019-07-01 00:00:00.926 | coinbase | BTC-USD | 2019-07-01 00:00:00.926 | 2019-07-01 00:00:01.444908 | False | bid | 10733.75 | NaN |
| 2019-07-01 00:00:00.926 | coinbase | BTC-USD | 2019-07-01 00:00:00.926 | 2019-07-01 00:00:01.444922 | False | bid | 10735.06 | NaN |
外れ値の確認¶
df.describe()
| price | amount | |
|---|---|---|
| count | 1.000000e+05 | 7.220700e+04 |
| mean | 1.334097e+05 | 3.626727e+00 |
| std | 2.695268e+07 | 1.442468e+02 |
| min | 1.000000e-02 | 4.000000e-08 |
| 25% | 1.068040e+04 | 1.367835e-01 |
| 50% | 1.072972e+04 | 7.000000e-01 |
| 75% | 1.075993e+04 | 1.431781e+00 |
| max | 8.385506e+09 | 2.961931e+04 |
px.box(df, x="price")
px.line(df.loc[:, "price"].resample("10s").mean())
前半のデータに外れ値があるため、除外します。
cut_df = df.loc["2019-07-01 00:02":, :]
px.line(cut_df, x="timestamp", y="price")
四分位値を算出して、外れ値を除外します。
lower, upper = cut_df.loc[:, "price"].quantile((0.005, 0.99))
filtered_df = cut_df.loc[
(cut_df.loc[:, "price"] > lower) & (cut_df.loc[:, "price"] < upper)
]
px.box(filtered_df, x="price")
df.describe()
| price | amount | |
|---|---|---|
| count | 1.000000e+05 | 7.220700e+04 |
| mean | 1.334097e+05 | 3.626727e+00 |
| std | 2.695268e+07 | 1.442468e+02 |
| min | 1.000000e-02 | 4.000000e-08 |
| 25% | 1.068040e+04 | 1.367835e-01 |
| 50% | 1.072972e+04 | 7.000000e-01 |
| 75% | 1.075993e+04 | 1.431781e+00 |
| max | 8.385506e+09 | 2.961931e+04 |
散布図に可視化します。
side = filtered_df.groupby("side")
bid, ask = side.get_group("bid"), side.get_group("ask")
fig = go.Figure(
[
go.Scattergl(
mode="markers",
x=bid.loc[:, "timestamp"],
y=bid.loc[:, "price"],
name="bid",
marker={
"color": np.log(bid.loc[:, "amount"]),
# "colorscale": "Blues",
},
),
go.Scattergl(
mode="markers",
x=ask.loc[:, "timestamp"],
y=ask.loc[:, "price"],
name="ask",
marker={
"color": -np.log(ask.loc[:, "amount"]),
# "colorscale": "Reds",
},
),
],
layout=go.Layout(
title="exchange: coinbase, instrument: btc-usd", width=1400, height=600
),
)
fig.show()