prophet

prophet1.py

import pandas as pd
import matplotlib.pyplot as plt
from fbprophet import Prophet
from datetime import datetime
import psycopg2
from sqlalchemy import create_engine

#Read the Parquet file
#df = pd.read_parquet('./BTCUSD-dec12-may21-volume-bars.parquet')

# read from db
alchemyEngine   = create_engine('postgresql+psycopg2://clj_user:bolzham_mojo@127.0.0.1/volumebars', pool_recycle=3600);
dbConnection    = alchemyEngine.connect();

pair = 'BTCUSD'
df = pd.read_sql(f'select sum(volume) as volume, min(price) as low, max(price) as high, max(time) as time from volume_bars where pair=\'{pair}\' group by group_cumsum', dbConnection);

# Select specific columns, high, low
selected_columns = df[['high', 'time']] 

# Convert timestamp in milliseconds to pandas timestamp
selected_columns['ds'] = pd.to_datetime(selected_columns['time'], unit='ms')

selected_columns = selected_columns.drop('time', axis=1)
selected_columns.columns = ['y', 'ds']

reordered_columns = ['ds', 'y']
df_reordered = selected_columns[reordered_columns]

# Show the resulting DataFrame
print(df_reordered.tail(5))


changepoints_df = pd.read_sql(f'select time from changepoints where pair =\'{pair}\' and spread > 777', dbConnection)

changepoints_df['ds'] = pd.to_datetime(changepoints_df['time'], unit='ms')
changepoints_df = changepoints_df.drop('time', axis=1)

changepoints = [x[0] for x in changepoints_df.values.tolist()]

model = Prophet(seasonality_mode='multiplicative', changepoint_prior_scale=1,changepoints=changepoints, changepoint_range=.9) 

model.fit(df_reordered)

future = model.make_future_dataframe(periods=8, freq='30min') #15 min interval #'h' is hourly
forecast = model.predict(future)

print(forecast[['ds', 'yhat_lower', 'yhat', 'yhat_upper', 'trend']].tail(8))
fig = model.plot(forecast, xlabel='Date', 
                 ylabel=r'yhat')

plt.title('price prediction - high')
plt.savefig(f"prophet-high-{datetime.now().strftime('%Y%m%d-%H%M%S')}.png")

xgboost.R

# Load the 'arrow' package
library(arrow)


#TODO: extract function from the code below
# Read data from a Parquet file
parquet_file <- "./BTCUSD-dec12-may21-volume-bars.parquet"

data <- arrow::read_parquet(parquet_file)

#make a data frame
df = as.data.frame(data) 
colnames(df)

#build features, indicators and target class
rsi = RSI(df$close, n=14, maType="WMA")
adx = data.frame(ADX(df[,c("high","low","close")]))
sar = SAR(df[,c("high","low")], accel = c(0.02, 0.2))
trend = df$Close - sar

#create a lag
rsi = c(NA,head(rsi,-1)) 
adx$ADX = c(NA,head(adx$ADX,-1))
trend = c(NA,head(trend,-1))
price = df$close-df$open

#target variable
class = ifelse(price > 0,1,0)

# Create a Matrix
model_df = data.frame(class,rsi,adx$ADX,trend)
model = matrix(c(class,rsi,adx$ADX,trend), nrow=length(class))
model = na.omit(model)
colnames(model) = c("class","rsi","adx","trend")

# Split data into train and test sets 
train_size = 2/3
breakpoint = nrow(model) * train_size

training_data = model[1:breakpoint,]
test_data = model[(breakpoint+1):nrow(model),]

# Split data training and test data into X and Y
X_train = training_data[,2:4] ; Y_train = training_data[,1]
class(X_train)[1]; class(Y_train)

X_test = test_data[,2:4] ; Y_test = test_data[,1]
class(X_test)[1]; class(Y_test)

# Train the xgboost model using the "xgboost" function
dtrain = xgb.DMatrix(data = X_train, label = Y_train)
xgModel = xgboost(data = dtrain, nround = 5, objective = "binary:logistic")

# Using cross validation
dtrain = xgb.DMatrix(data = X_train, label = Y_train)
cv = xgb.cv(data = dtrain, nround = 10, nfold = 5, objective = "binary:logistic")

# Make the predictions on the test data
preds = predict(xgModel, X_test)

# Determine the size of the prediction vector
print(length(preds))

# Limit display of predictions to the first 6
print(head(preds))

# Measuring model performance
error_value = mean(as.numeric(preds > 0.5) != Y_test)
print(paste("test-error=", error_value))

prediction = as.numeric(preds > 0.5)
print(head(prediction))

# View feature importance from the learnt model
importance_matrix = xgb.importance(model = xgModel)
print(importance_matrix)

#TODO: set device to print plots to png
# View the trees from a model
# xgb.plot.tree(model = xgModel)
# View only the first tree in the XGBoost model
# xgb.plot.tree(model = xgModel, n_first_tree = 1)

#how to save a checkpoint:
with open('volume-bars-model.json', 'w') as file_out:
    json.dump(model_to_json(model), file_out)

#how to load checkpoint
with open('volume-bars-model.json', 'r') as file_in:
    model = model_from_json(json.load(file_in))

then we just use it:

forecast = model.predict()
fig = model.plot(forecast)

#re-train Prophet from existing checkpoint:
import numpy as np
class StanInit:
    def init(self, model):
        self.params = {
            'k': np.mean(model.params['k']),
            'm': np.mean(model.params['m']),
            'sigma_obs':
             np.mean(model.params['sigma_obs']),
            'delta': np.mean(model.params['delta'],
                             axis=0),
            'beta': np.mean(model.params['beta'], axis=0)
        }
    def call(self):
        return self.params

fine-tuning

model2 = Prophet().fit(df, init=StanInit(prev_model))

#use plotly

from fbprophet.plot import (
    plot_plotly,
    plot_components_plotly,
    plot_forecast_component_plotly,
    plot_seasonality_plotly
)

fig = plot_plotly(model, forecast, trend=True)
py.iplot(fig)

#components
fig = plot_components_plotly(model, forecast ,
                             figsize=(800, 175))
py.iplot(fig)

#seasonality
fig = plot_seasonality_plotly(model, 'yearly')
py.iplot(fig)