creating new features with certain percentile of price

Question

I am working on a forex classification problem, need help with creating the below-detailed features, I have shared my code below and also attached pic for a visual reference of the issue at hand.

Feature: opensimilarclose (1 if open = close plus or minus 2 pips, 0 otherwise)

Feature: opencloselow (1 if both open and close > 90% of candle size, 0 otherwise)

Feature: openclosehigh (1 if both open and close < 10% of candle size, 0 otherwise)

MY CODE:

data['opensimilarclose'] = np.where(data.Open-data.Close<=0.02, 1,0)

data['openclosehigh'] = np.where((abs(data.Close-data.Low)>=abs(data.High-data.Low)*0.9 and ()), 1, 0)

data['opencloselow'] = np.where(abs(data.Close-data.Low)<=abs(data.High-data.Low)*0.1, 1, 0)

please find sample of the data below:

Date    Timestamp   Open    High    Low Close   Volume
2004-01-01  00:00:00    414.92199999999997  414.92199999999997  414.23199999999997  414.55800000000005  0.738269000896253
2004-01-02  00:00:00    414.32199999999995  416.098 413.86699999999996  415.395 3.82642700810902
2004-01-04  00:00:00    414.278 414.69800000000004  414.096 414.444 0.0564850000591832
2004-01-05  00:00:00    415.376 423.981 414.23400000000004  421.89300000000003  10.4188560213806
2004-01-06  00:00:00    422.332 430.17800000000005  420.07800000000003  421.777 11.182643023759699
2004-01-07  00:00:00    420.773 424.121 418.974 419.626 11.956311026187901
2004-01-08  00:00:00    419.574 424.798 416.27  423.298 12.439296027514501
2004-01-09  00:00:00    423.298 426.897 419.42699999999996  425.404 9.2499640192309
2004-01-11  00:00:00    426.49800000000005  426.49800000000005  425.876 426.23  0.0673800002332428
2004-01-12  00:00:00    425.853 428.459 422.219 424.598 10.6995250192995
2004-01-13  00:00:00    424.598 426.395 421.651 423.69800000000004  11.1990780260712
2004-01-14  00:00:00    423.389 424.397 416.78  419.298 10.835633025399101
2004-01-15  00:00:00    418.98  421.098 406.906 408.44699999999995  12.266192030985598
2004-01-16  00:00:00    408.546 410.398 404.43300000000005  406.298 9.26100601695725
2004-01-18  00:00:00    405.842 406.098 405.543 405.75300000000004  0.0658050001220545
2004-01-19  00:00:00    407.18800000000005  408.68300000000005  405.402 406.751 5.688531011830491
2004-01-20  00:00:00    406.449 412.69699999999995  404.417 411.921 10.6885030245794
2004-01-21  00:00:00    411.99800000000005  412.91  406.721 409.832 10.672994028404
2004-01-22  00:00:00    410.043 412.69800000000004  407.216 409.033 9.949593026152801
2004-01-23  00:00:00    409.398 412.29699999999997  405.461 407.398 8.921345019130971

Answer 1

You have few small errors in your code:

1. You check only if Open-Close is smaller then 0.02, and forget to check for absolute value (if open=5 and close=8 and still smaller then 0.02)
2. "openclosehigh" and "opencloselow" are different in your code from what you say they are suppose to be. To take into consideration only close price.

I personally prefer to work with pandas directly instead of where since it's unneeded - you have a simple condition.

Check the following example:

import pandas as pd

df = pd.DataFrame({"Open": [4, 3.6, 7, 6], "Close": [4.1, 3.5, 6.7, 6.8], "High": [4.12, 3.6, 7.02, 6.8], "Low":[4, 3.498, 6.7, 5.7]})
df["opensimilarclose"] = (abs(df["Open"] - df["Close"]) <= 0.02).astype(int)
df["relative_open"] = (df["Open"] - df["Low"]) / (df["High"] - df["Low"]) 
df["relative_close"] = (df["Close"] - df["Low"]) / (df["High"] - df["Low"]) 

df["openclosehigh"] = ((df["relative_open"] > 0.9) & (df["relative_close"] > 0.9)).astype(int)
df["opencloselow"] = ((df["relative_open"] < 0.1) & (df["relative_close"] < 0.1)).astype(int)

The 3rd line calculating opensimiliarclose by directly asking if the absolute different between open and close is smaller then 0.02. It's a condition, so the result is True/False. To change to 1/0 I added .astype(int). This formatting of directly applying condition over all column is more convenient in my opinion then using where.

Then for your second and third columns, I though it's more convenient to first calculate the percentage and then check the condition. The column "relative_open" and "relative_close" holds the percentages of open/close, and only in the next two lines I condition on both to fill "opencloselow" and "openclosehigh". You can remove the extra columns by drop or loc over all other columns. You can also just put the result as a temporary series instead of extra column (tmp_series = (df["Close"]...).

Answer 2

Investing Better you can visualize the candle using the following code...you can see red and green candles...the color of the candle is determined by previous close...but here i did not use previous close...

import matplotlib.pyplot as plt
import plotly.graph_objects as go

fig = go.Figure(data=[go.Candlestick(x=df.index,
            open=df['Open'],
            high=df['High'],
            low=df['Low'],
            close=df['Close'])],
            layout={'height':500,'width':1000})
fig.update_layout(xaxis_rangeslider_visible=False)

fig.show()

this is industry standard. I understand you take previous close and calculate other features....