Masoud-Khosravi / Machine-Learning

Review of some machine learnings models

This is a tutorial for $\color{#1DADFF}Machine \space Leraning$ algorithms

$\color{red}Hello \space \color{lightblue}and \space \color{orange}Wellcome$

$\color{red}WARNING$ : DO NOT USE THESE IN REAL MARKET

In this Package we try input data from stock markets and evaluate some of the machine learnings models

(for Now and later maybe we add other options )

Using this Package is too easy:

• Befor use you should know--> how can read and input data

In this Learnings we study some models of supervised learning

• Classification
• Regression

we study this models in Classification

• TensorFlow
• Random Forest
• Ada Boost
• K-Nearest Neighbors
• Decision Trees
• Support Vector Machines

And we study this models in Regression

• TensorFlow
• Support Vector Regression
• Decision Tree Regression
• ridge Regression

and in here I put some codes to show how its work:

• Classification
• Regression

Now --> The first step is -> Read the data

We have already made this function

Read More: Stock Markets

Example:

from ReadData import MetaTrader
df_raw=my_obj.df_type1_raw
df_raw.tail()

Example for Classification: it's time to make y_values

at this part we create the y values

we calculate stop loss in each candles close price based on low26/high26

<low26: for buy signals & high26: for sell signals>

and we set take profit at least twice (means TP>=2*SL)

Then we give time to the program: in this section time is equal to 24 candles

Means: this model must take profit before reaching the stop loss in maximum 24 candles

So

• if it is profitable in sell means our signal should be 'Sell'
• if it is profitable in buy means our signal should be 'Buy'
• if the model does not take profit in any of them before reaching the stop loss --> our signal is 'Nothing' in that candles close time

we show:

• buy signals by->1
• sell signals by->2
• and Do Nothings by->0

lets go

def create_y(dataframe):
    """
    Please pass your raw dataframe
    """
    df=dataframe.copy()
    #create an empty y values
    y=pd.DataFrame(np.nan,index=np.arange(len(df)),columns=["y_true"])
    
    # in this part from time the only thing that matters to us is the hour
    df['time'] = df['time'].apply(lambda num: num.hour)
    for pos in np.arange(0, len(df)):
        close_price = df.loc[pos, 'close']
        # calculate distance of stop loss and create tp points
        sl_buy = close_price-df.loc[pos, 'Low26']
        sl_sell = df.loc[pos, 'High26']- close_price
        
        tp_buy = sl_buy * 2.1
        tp_sell = sl_sell * 2.1
        signal = 0 # default signal is Zero (do nothing)        
        # we don't trade after 11 p.m until 4 a.m menas in this hour signals should be Zero
        if (df.loc[pos, 'time'] < 23) and (df.loc[pos, 'time'] > 4):
            for last in np.arange(1, 25):# This is the part where we give maximum time to the model: 24 candles
                if (pos + last) < len(df):
                    high_last = df.loc[pos + last, 'High26'] - close_price
                    low_last = close_price- df.loc[pos + last, 'Low26']
                    low_now = close_price- df.loc[pos + last, 'low']
                    high_now = df.loc[pos + last, 'high'] - close_price
                    if (low_last <= sl_buy) and (high_now > tp_buy):
                        signal = 1
                        break
                    elif (high_last <= sl_sell) and (low_now > tp_sell):
                        signal = 2
                        break
        y.loc[pos] = signal
    return y

y=create_y(df_raw)

from Learnings import Classifications

After we create y we pass normalized dataframe to models

So we have normalized df:

df = my_obj.df_type1_changed
y = pd.to_numeric(y['y_true'], downcast='integer')
Class_obj = Classifications(df,y)

Let's check the models:

model = Class_obj.tensorflow()
Class_obj.evaluates.plot_confusion_matrix()

model = Class_obj.random_forest()
Class_obj.evaluates.plot_confusion_matrix()

As you can see we have one new row in evaluation!

Yes that is So What is it ?

As you know

• when our model predict 0 mean no signal --> We don't lose money AND We don't earn money
• when our model predict 1 or 2 --> We have Buy/Sell signal
  • if we loss --> we lose money by stop loss (SL)
  • if we won --> we earn money by take profit (TP>2*SL)

So --> How Calculate final Score ?

• 0-> No Signal -> No Point
• 1/2 (Buy/Sell) if it was True --> We GET 2 Point (2*SL)
• 1/2 (Buy/Sell) if it was False --> We LOSE 1 point (SL)
  • Final Score= True Signals*2 - False Signal

Now let's see an example for Regression:

after reading data we should make y_values

in this part we create the y values

our y values in this case is The closing price of the next candle

and for best selection we use distance of next close price vs current close price

this_close - next_close --> So

• if distance is positive means next candle is lower than this candle As long as the distance
• if distance is negative means next candle is upper than this candle As long as the distance

lets go

def create_y(dataframe):
    """
    Please pass your raw dataframe
    """
    df=dataframe.copy()
    y = df.apply(lambda row: df["close"][row.name]-df["close"][row.name+1] 
                                if row.name<(len(df)-1) else np.nan,axis=1)
    return y

now we can visulize data

after cleaning and organizing dataframe we plot again:

Let's check some models:

from Learnings import Regression
model = Class_obj.tensorflow()
Class_obj.evaluates.mean_errors()

Class_obj.evaluates.plot_df_y()

and we can see Some or All predicts:

df_y[df_y['y_pred']<=-2]

as you can See it's not bad and you can see all steps here:

• Classification
• Regression

And Final Note :

Never use this functions for Trade, These are only educational

under development ... < we will come back soon >

Other Repositories:

$\color{red}About \space \color{orange}Me:$