Skip to content

Instantly share code, notes, and snippets.

@j40903272

j40903272/evaluation.py

Created February 10, 2022 05:00
Show Gist options
  • Save j40903272/67a1beb6c5138fd44cafec5b5897f025 to your computer and use it in GitHub Desktop.
Save j40903272/67a1beb6c5138fd44cafec5b5897f025 to your computer and use it in GitHub Desktop.
Download ZIP
riusbot 近期績效整理
Raw
evaluation.py
#!/usr/bin/env python
# coding: utf-8
# In[59]:
import pdb
import time
import pickle
import ccxt
import requests
import logging
import functools
import datetime
import pandas as pd
import numpy as np
from tqdm import tqdm
from collections import defaultdict
import os
import concurrent.futures
import time
import pandas as pd
import matplotlib.pyplot as plt
get_ipython().run_line_magic('matplotlib', 'inline')
# In[270]:
with open("/tmp/message.pkl", "rb") as f:
message = pickle.load(f)
# In[3]:
exchange = ccxt.binance({
'enableRateLimit': True,
'options': {
'defaultType': 'spot',
},
})
tmp = exchange.loadMarkets(True)
# In[263]:
class Evaluation():
def __init__(self, base, hold, leverage=1, timeframe="8h", sl=0.0, tp=0.0, move=99999):
self.leverage = leverage
self.base = base
self.hold = hold # minutes
self.total_profit = 0
self.timeframe = timeframe
self.sl = sl
self.tp = tp
self.config = {
"sl": sl,
"tp": tp,
"leverage": leverage,
}
"klines : time 0, open 1, high 2, low 3, close 4, vol 5"
def evaluate(self, info: dict, symbol: str):
since = int(info["timestamp"] * 1000)
limit = self.hold
klines = exchange.fetch_ohlcv(
symbol, timeframe=self.timeframe, since=since, limit=limit
)
entry = klines[0][1]
action = info["action"]
entry_timestamp = float(klines[0][0]) / 1000
liquidate = entry * (1 + 1 / self.leverage)
profit = 0
result = None
for e, k in enumerate(klines):
low = float(k[3])
high = float(k[2])
close_timestamp = float(k[0])/1000
# if high >= liquidate:
# profit = -self.base
# percentage = -1
# close = liquidate
# result = "liquidate"
# break
if action == "SELL":
if self.sl:
sl = entry * (1+self.sl) if info["sl"] is None else info["sl"]
if high >= sl:
close = sl
if action == "SELL":
percentage = -(close / entry - 1)
else:
percentage = (close / entry - 1)
profit = percentage * self.base
result = "stop loss"
break
if self.tp:
tp = entry * (1-self.tp) if info["tp"] is None else info["tp"]
if low <= tp:
close = tp
if action == "SELL":
percentage = -(close / entry - 1)
else:
percentage = (close / entry - 1)
profit = percentage * self.base
result = "take profit"
break
else:
if self.sl:
sl = entry * (1-self.sl) if info["sl"] is None else info["sl"]
if low <= entry * (1-self.sl):
close = sl
if action == "SELL":
percentage = -(close / entry - 1)
else:
percentage = (close / entry - 1)
profit = percentage * self.base
result = "stop loss"
break
if self.tp:
tp = entry * (1+self.tp) if info["tp"] is None else info["tp"]
if high >= tp:
close = tp
if action == "SELL":
percentage = -(close / entry - 1)
else:
percentage = (close / entry - 1)
profit = percentage * self.base
result = "take profit"
break
if profit == 0:
close = float(k[4])
if action == "SELL":
percentage = -(close / entry - 1)
else:
percentage = (close / entry - 1)
profit = percentage * self.base * self.leverage
result = "end"
if "action" not in info:
print(info)
return {
"symbol": symbol,
"date": info["date"],
"action": info["action"],
"entry_timestamp": int(entry_timestamp),
"close_timestamp": int(close_timestamp),
"entry": entry,
"close": close,
"liquidate": liquidate,
"profit": profit,
"percentage": percentage,
"result": result,
"klines": klines,
}
def simulate(self, trades):
detail = list()
tasks = list()
with concurrent.futures.ThreadPoolExecutor(max_workers=10) as executor:
for order in trades:
# if order["action"] != "sell":
# continue
symbol = f'{order["symbol"]}/USDT'
try:
exchange.market(symbol)
except:
print(symbol, "not found")
continue
info = {
"date": order['date'],
"timestamp": order['date'].timestamp(),
"action": order["action"],
"sl": order["stop_loss"],
"tp": order["take_profit"],
}
task = executor.submit(self.evaluate, info, symbol)
tasks.append(task)
time.sleep(0.05)
for task in tqdm(concurrent.futures.as_completed(tasks), total=len(tasks)):
result = task.result()
if result:
detail.append(result)
self.total_profit += result["profit"]
self.detail = detail
daily_profit = np.array([i["percentage"] for i in s.detail])
plt.hist(daily_profit, bins=50)
daily_profit *= self.leverage
self.config["profit"] = self.total_profit
self.config["mean"] = np.mean(daily_profit)
self.config["std"] = np.std(daily_profit)
self.config["sharpe"] = (np.mean(daily_profit) - 0.0005) / np.std(daily_profit)
self.save_records()
return self.config
def save_records(self, path="evaluation.csv"):
if os.path.exists(path):
df = pd.read_csv(path)
else:
df = pd.DataFrame(columns=self.config.keys())
df = df.append(self.config, ignore_index=True)
df.to_csv(path, index=False)
def output_df(self):
columns = list(self.detail[0].keys())
df = pd.DataFrame()
for col in columns:
df[col] = [i[col] for i in self.detail]
return df
# # ROSE
# In[197]:
rose_trade = []
for i in message:
i["date"] = datetime.datetime.fromtimestamp(i["message_timestamp"])
i["timestamp"] = int(i["message_timestamp"])
if i["channel"] == "ROSE":
rose_trade.append(i)
rose_trade = rose_trade[:-40][::-1]
print(len(rose_trade))
# In[219]:
s = Evaluation(base=100, hold=48, timeframe="1h", leverage=1, sl=0.15, tp=0.35)
result = s.simulate(rose_trade)
print(result)
df = s.output_df()
df["day"] = df.date.apply(lambda x: x.strftime("%Y%m%d"))
tmp = df.groupby("day").sum("profit").percentage.values
tmp *= 3
print((np.mean(tmp) - 0.0004) / np.std(tmp))
display(df.sort_values("date", ascending=True))
# In[225]:
df = df.sort_values("date", ascending=True)
init_balance = 1000
x = df.date
y = np.cumsum(df.profit.values.tolist()) + init_balance
from scipy.ndimage.filters import gaussian_filter1d
y = gaussian_filter1d(y, sigma=10)
plt.plot(x, y)
plt.xticks(rotation=45)
# In[221]:
init = 100
balance = []
for i in df.percentage.values:
init *= (i+1)
balance.append(init)
x = df.date
y = balance
from scipy.ndimage.filters import gaussian_filter1d
y = gaussian_filter1d(y, sigma=10)
plt.plot(x, y)
plt.xticks(rotation=45)
# In[222]:
df.to_csv("rose_20220209.csv", index=False)
# # Perpetual
# In[226]:
perp_trade = []
for i in message:
i["date"] = datetime.datetime.fromtimestamp(i["message_timestamp"])
i["timestamp"] = int(i["message_timestamp"])
if i["symbol"] in "BTCUSDT_22":
i["symbol"] = "BTC"
if i["symbol"] in "ETHUSDT_22":
i["symbol"] = "ETH"
if i["channel"] == "PERPETUAL":
perp_trade.append(i)
print(len(perp_trade))
# In[233]:
s = Evaluation(base=100, hold=8, timeframe="1h", leverage=1, sl=0.2, tp=0.2)
result = s.simulate(perp_trade)
print(result)
df = s.output_df()
df["day"] = df.date.apply(lambda x: x.strftime("%Y%m%d"))
tmp = df.groupby("day").sum("profit").percentage.values
tmp *= 3
print((np.mean(tmp) - 0.0004) / np.std(tmp))
display(df.sort_values("date", ascending=True))
# In[234]:
df = df.sort_values("date", ascending=True)
df.to_csv("perpetual_20220209.csv", index=False)
init_balance = 1000
x = df.date
y = np.cumsum(df.profit.values.tolist()) + init_balance
from scipy.ndimage.filters import gaussian_filter1d
y = gaussian_filter1d(y, sigma=1)
plt.plot(x, y)
plt.xticks(rotation=45)
# In[235]:
init = 100
balance = []
for i in df.percentage.values:
init *= (i+1)
balance.append(init)
x = df.date
y = balance
from scipy.ndimage.filters import gaussian_filter1d
y = gaussian_filter1d(y, sigma=1)
plt.plot(x, y)
plt.xticks(rotation=45)
# # Daily
# In[288]:
daily_trade = []
for i in message:
i["date"] = datetime.datetime.fromtimestamp(i["message_timestamp"])
i["timestamp"] = int(i["message_timestamp"])
if i["symbol"] in "BTCUSDT_22":
i["symbol"] = "BTC"
if i["symbol"] in "ETHUSDT_22":
i["symbol"] = "ETH"
if "KSM" in i["symbol"]:
continue
# i["stop_loss"] = None
# i["take_profit"] = None
if i["channel"] == "DAILYSCALP":
daily_trade.append(i)
print(len(daily_trade))
# In[293]:
s = Evaluation(base=100, hold=108, timeframe="1h", leverage=1, sl=0.2, tp=0.4)
result = s.simulate(daily_trade)
print(result)
df = s.output_df()
df["day"] = df.date.apply(lambda x: x.strftime("%Y%m%d"))
tmp = df.groupby("day").sum("profit").percentage.values
tmp *= 3
print((np.mean(tmp) - 0.0004) / np.std(tmp))
display(df.sort_values("date", ascending=True))
# In[295]:
df = df.sort_values("date", ascending=True)
df.to_csv("daily_20220209.csv", index=False)
init_balance = 1000
x = df.date
y = np.cumsum(df.profit.values.tolist()) + init_balance
from scipy.ndimage.filters import gaussian_filter1d
y = gaussian_filter1d(y, sigma=1)
plt.plot(x, y)
plt.xticks(rotation=45)
# # Justin
# In[297]:
just_trade = []
for i in message:
i["date"] = datetime.datetime.fromtimestamp(i["message_timestamp"])
i["timestamp"] = int(i["message_timestamp"])
if i["symbol"] in "BTCUSDT_22":
i["symbol"] = "BTC"
if i["symbol"] in "ETHUSDT_22":
i["symbol"] = "ETH"
# i["stop_loss"] = None
# i["take_profit"] = None
if i["channel"] == "JUSTIN":
just_trade.append(i)
print(len(just_trade))
# In[307]:
s = Evaluation(base=100, hold=120, timeframe="1h", leverage=1, sl=0.2, tp=0.2)
result = s.simulate(just_trade)
print(result)
df = s.output_df()
df["day"] = df.date.apply(lambda x: x.strftime("%Y%m%d"))
tmp = df.groupby("day").sum("profit").percentage.values
tmp *= 3
print((np.mean(tmp) - 0.0004) / np.std(tmp))
display(df.sort_values("date", ascending=True))
# In[309]:
df = df.sort_values("date", ascending=True)
df.to_csv("justin_20220209.csv", index=False)
init_balance = 1000
x = df.date
y = np.cumsum(df.profit.values.tolist()) + init_balance
from scipy.ndimage.filters import gaussian_filter1d
y = gaussian_filter1d(y, sigma=10)
plt.plot(x, y)
plt.xticks(rotation=45)
# # Vegas
# In[311]:
vegas_trade = []
for i in message:
i["date"] = datetime.datetime.fromtimestamp(i["message_timestamp"])
i["timestamp"] = int(i["message_timestamp"])
if i["symbol"] in "BTCUSDT_22":
i["symbol"] = "BTC"
if i["symbol"] in "ETHUSDT_22":
i["symbol"] = "ETH"
# i["stop_loss"] = None
# i["take_profit"] = None
if i["channel"] == "VEGAS":
vegas_trade.append(i)
print(len(vegas_trade))
# In[315]:
s = Evaluation(base=100, hold=48, timeframe="1h", leverage=1, sl=0.1, tp=0.3)
result = s.simulate(vegas_trade)
print(result)
df = s.output_df()
df["day"] = df.date.apply(lambda x: x.strftime("%Y%m%d"))
tmp = df.groupby("day").sum("profit").percentage.values
tmp *= 3
print((np.mean(tmp) - 0.0004) / np.std(tmp))
df = df.sort_values("date", ascending=True)
display(df)
# In[318]:
df = df.sort_values("date", ascending=True)
df.to_csv("vegas_20220209.csv", index=False)
init_balance = 1000
x = df.date.values[2:]
y = np.cumsum(df.profit.values.tolist()[2:]) + init_balance
from scipy.ndimage.filters import gaussian_filter1d
y = gaussian_filter1d(y, sigma=1)
plt.plot(x, y)
plt.xticks(rotation=45)
# In[ ]:
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment