Spaces:
Runtime error
Runtime error
import pytz | |
import datetime | |
import io | |
import pandas as pd | |
def time_in_beijing(): | |
''' | |
return current time in Beijing as datetime object | |
''' | |
tz = pytz.timezone('Asia/Shanghai') | |
dt = datetime.datetime.now(tz) | |
return dt | |
def convert_string_to_datetime(date_string, time_zone="Asia/Shanghai"): | |
''' | |
Convert a string to a datetime object with the timezone by default, | |
Shanghai | |
''' | |
dt = datetime.datetime.strptime(date_string, '%Y-%m-%d %H:%M:%S.%f') | |
tz = pytz.timezone(time_zone) | |
dt = tz.localize(dt) | |
return dt | |
def create_stocks_entry_from_excel(byte_string): | |
'''create stock entry from excel file | |
Parameters | |
---------- | |
byte_string: bytes | |
the byte string of the excel file | |
Returns | |
------- | |
new_stock_entry: list | |
[{ticker:str, shares:int, mean_price: float, date:datetime.datetime}] | |
the list of stock entry | |
''' | |
uploaded_df = None | |
with io.BytesIO(byte_string) as f: | |
uploaded_df = pd.read_excel(f) | |
# throw exception if doesn't have required columns | |
if not set(['证券代码', '持仓数量', '平均建仓成本', 'time_stamp']).issubset(uploaded_df.columns): | |
raise Exception('Missing required columns') | |
# print(uploaded_df) | |
# uploaded_df = pd.read_excel() | |
uploaded_df.drop(columns='Unnamed: 0', inplace=True) | |
# Define the regular expression pattern to match the string endings | |
pattern = r'\.(sz|sh)$' | |
# Define the replacement strings for each match group | |
replacements = {'.sz': '.XSHE', '.sh': '.XSHG'} | |
# Use the str.replace method with the pattern and replacements | |
uploaded_df['证券代码'] = uploaded_df['证券代码'].str.lower() | |
uploaded_df['证券代码'] = uploaded_df['证券代码'].str.replace( | |
pattern, lambda m: replacements[m.group()], regex=True) | |
new_stock_entry = [ | |
dict(ticker=ticker, shares=shares, date=time, mean_price=mean_price) | |
for ticker, shares, mean_price, time in zip( | |
uploaded_df['证券代码'], | |
uploaded_df['持仓数量'], | |
uploaded_df['平均建仓成本'], | |
pd.to_datetime(uploaded_df['time_stamp']))] | |
# new_profile, error = api.update_portfolio_profile(new_stock_entry) | |
print(new_stock_entry) | |
return new_stock_entry | |
def style_number(vals): | |
'''color negative number as red, positive as green | |
Parameters | |
---------- | |
vals: df columns | |
the columns to be styled | |
Returns | |
------- | |
list | |
the list of style | |
''' | |
return ['color: red' if v < 0 else 'color: green' for v in vals] | |
def create_share_changes_report(df): | |
'''Create a markdown report of the share changes for certain date | |
Parameters | |
---------- | |
df: pd.DataFrame | |
the dataframe of profile for a specific date | |
Returns | |
------- | |
markdown: str | |
''' | |
date_str = df.date.to_list()[0].strftime('%Y-%m-%d %H:%M:%S') | |
markdown = f"### {date_str}\n\n" | |
markdown += 'Ticker | Display Name | Share Changes\n' | |
markdown += '--- | --- | ---\n' | |
for _, row in df.iterrows(): | |
share_changes = row['share_changes'] | |
# Apply green color to positive numbers and red color to negative numbers | |
if share_changes > 0: | |
share_changes_str = f'<span style="color:green">{share_changes}</span>' | |
elif share_changes < 0: | |
share_changes_str = f'<span style="color:red">{share_changes}</span>' | |
else: | |
share_changes_str = str(share_changes) | |
markdown += '{} | {} | {}\n'.format(row['ticker'], | |
row['display_name'], share_changes_str) | |
return markdown | |