README.md

Tradeformer

-Transformer for trading

Tokenizer

Built a BPE style tokenizer for markets. The idea is to convert each candle into a character depending on its properties. Below are the properties used to convert a candle into a character. Then run BPE to generate a vocab and merges file.

Candle to Chars

```
G = Bull Candle = A
R = Bear Candle = B
N = No Color Candle = C

E = Candle == ATR ratio = D
SSS = Candle <<< ATR = E
SS = Candle << ATR = F
S = Candle < ATR = G
BBB = Candle >>> ATR = H
BB = Canadle >> ATR = I
B = Candle > ATR = J

M = Top Wick == Bot Wick = K
LLL = Bot Wick >>> TopWick = L
LL = Bot Wick >> TopWick = M 
L = Bot Wick > TopWick = N

UUU = Top Wick >>> Bot Wick = O
UU = Top Wick >> Bot Wick = P 
U = Top Wick > Bot Wick = Q

```

Model

The model used is a causal LM (GPT like model). Total number of parameters in 10 million.