Talib RSI wrong values

[josuhap]

I'm developing a small python script to get the RSI of a stock using TA-Lib. It was working fine for 6 months but now I realized the RSI function from TA-Lib is returning worng values:

while True: symbol = 'BTCUSDT' for period in periods: try: print(f'{symbol} - {period}') diff = ['1','2','3'] for hour in diff: ago = f'{hour} hours ago' datas = client.get_historical_klines(symbol,period, ago) for data in datas[:-1]: timestamp = converttime(data[0], 'full') price = round(float(data[4]),2) closes.append(price) np_closes = np.array(closes) erre = talib.RSI(np_closes, 14) rsis = erre.tolist() print(f'{timestamp} -> {symbol} -> {rsis[-1]} - {ago}')

Here an screenshot trying to get RSI from different timeframes (1m and 5m) with the difference of the candle numbers.

I have a big question... If RSI function only gets last 14 values, why the value is so different between those tfs?

Correct value at 5m tf

Correct value at 1m tf

Version:

import talib talib.__ta_version__ b'0.4.0 (Oct 16 2019 22:53:18)'

from talib import abstract stoch = abstract.Function('stoch') stoch.set_parameters(fastk_period=14, slowk_period=1, slowd_period=3, slowk_matype=0, slowd_matype=0) stoch.lookback 15

[trufanov-nok]

You may better read the discussion about indicators with "memory effect" here: https://github.com/mrjbq7/ta-lib/issues/469

[josuhap]

Hello @trufanov-nok,

First of all thank you for your response, I have read that discussion but I think the problem is not the same.

To avoid "memory effect" about indicators, I use this commands:

closes = 0 np_closes = 0 erre = 0 rsis = 0 datas = 0

So I am assuming that with those commands I "reset" all the values of them.

In none of the cases the RSI is the correct (1m -> 1/2/3 hours ago and 5m -> 1/2/3 hours ago).

PD: I understand the "memory effect" but only I need is the last value rsis[-1], and that's a worng value

[mrjbq7]

I think you misunderstand "memory effect", it's not the memory used by Python objects, but rather the influence on your latest RSI value by all previous observed values.

[trufanov-nok]

I reproduced this code and assume the problem is in the difference between data returned to python via binance API and data displayed on graph online. I mean their klines' closes are different. Sometimes significantly. Could you confirm?

For the reference my test server code is:

from datetime import datetime
import numpy as np
import talib
from binance.client import Client
client = Client("ALUdiXPFeGbBRa533X8msQJNxp75st0F7W98uInhAg6pGFXD5OXcuDiLwkgjbRxy", "gl5Mcg9CnLOEzYWwN0z2hFgWSySsf3sI8AxzIOsDbItyvcGrk9cASmGxfDrwK3BF", testnet=True)

symbol = 'BTCUSDT'
periods = [Client.KLINE_INTERVAL_5MINUTE, Client.KLINE_INTERVAL_1MINUTE]
diff = ['1','2','3', '10']
closes = []

for period in periods:
    print(f'{symbol} - {period}')
    for hour in diff:
        ago = f'{hour} hours ago'
        datas = client.get_historical_klines(symbol,period, ago)
        for data in datas[:-1]: 
            timestamp = datetime.utcfromtimestamp(int(data[0]/1000)).strftime('%Y-%m-%d %H:%M:%S')
            price = round(float(data[4]),2)
            closes.append(price)
        np_closes = np.array(closes)

        erre = talib.RSI(np_closes, 14)
        rsis = erre.tolist()
        print(f'{timestamp} -> {symbol} -> {rsis[-1]} - {ago}')

[hossain93]

you must have a big data because stock calculate rsi from the first date but you use a specefic date, and this diffrent is effective on your calculate but if your date was long the diffrent is small sorry for my weak english:)