Render financial data

For now there is only one renderer, the TensorboardDataRenderer. It helps you to render financial data and results of machine learning training to Tensorboard.

TensorboardDataRenderer

** Example: **

from datetime import datetime
import pandas as pd
from Hmile.DataRenderer import TensorboardDataRenderer
from Hmile.DataProvider import CSVDataProvider
from Hmile.FillPolicy import FillPolicyAkima
PAIR = "BTCUSD"
START = "2021-12-01"
END = "2022-12-31"
DATA_DIR = "/my/data/dir"
INTERVAL = "hour"

def fill_renderer(data, renderer):
# we fill the renderer with data rows
for index, row in data.iterrows():
    renderer.append("open", row["open"], index)
    renderer.append("close", row["close"], index)
    renderer.append("high", row["high"], index)
    renderer.append("low", row["low"], index)
    renderer.append("volume", row["volume"], index)

# we create a renderer object
renderer = TensorboardDataRenderer('logs/')
# we create a data provider object
dp = CSVDataProvider([PAIR], START, END, DATA_DIR, interval=INTERVAL)
# we set a fill policy
dp.fill_policy = FillPolicyAkima(INTERVAL)
# we read data
data = dp.getData()[PAIR]
# we fill renderer
fill_renderer(data, renderer)
# we launch renderer
renderer.render()
# then we increment tensorboard step
renderer.next_step()
# we refill the renderer
fill_renderer(data, renderer)
# we launch renderer
renderer.render()

Exemple result in tensorboard :

Remark :

In the upper example we only used financial data. However you can add model training info.

printables informations

info name	required	description
open	yes	open price
high	yes	high price
close	yes	close price
volume	yes	volume price
money	no	actual possessed money
rew	no	actual training reward
short	no	did the model returns a short signal signal ?
long	no	did the model returns a long signal signal ?
exit	no	did the model returns a exit signal signal ?

So you could add any information you want by adding a new key to the renderer in the fill_renderer loop. As :

renderer.append("rew", random.random()*1000, index)
renderer.append("long", random.choose([1, None]), index)