performance.py

"""
This module calculates different performance metrics for the 
backtests and saves them in a summary folder
"""
import os
import json
import tempfile
import concurrent.futures
import pandas as pd
import numpy as np
import pyfolio as pf
from fastbt.rapid import metrics

try:
    from google.cloud import storage
    GOOGLE_CLOUD = True
    client = storage.Client()
except ImportError:
    print('Google Cloud library not installed')
    GOOGLE_CLOUD = False

def get_benchmark(filename='indices.csv', parse_dates=['Date']):
    """
    Get the benchmark returns
    filename
        filename with date and ohlc columns
    parse_dates
        columns to parse dates as list
    returns calculated on the close column
    """
    df = pd.read_csv(filename, parse_dates=parse_dates)
    df.rename(lambda x: x.lower(), axis='columns', inplace=True)
    df = df.set_index('date').sort_index()
    df['chg'] = df.close.pct_change(1)
    df = df.loc['2012':]
    return df

def all_metrics(result, benchmark):
    """
    Calculate all the metrics for the given backtest results
    result
        backtest results in the expected format - results generated
        by rapid
    """
    dct = {}
    byday = result.groupby('timestamp').net_profit.sum()
    stats = pf.timeseries.perf_stats(byday/100000,
        factor_returns=benchmark.chg)
    dct.update(stats.to_dict())
    simple = metrics(result)
    dct.update(simple)
    dct['open_high'] = result.query('open==high').net_profit.sum()
    dct['open_low'] = result.query('open==low').net_profit.sum()
    by_year = byday.groupby(lambda x: x.year).sum()
    dct.update(by_year.to_dict())
    return dct

def runner(filename, output_file, benchmark, counter):
    """
    Runner for concurrent execution
    """
    results = pd.read_hdf(filename)
    perf_stats = all_metrics(results, benchmark)
    print(counter, output_file)
    with open(output_file, 'w') as f:
        json.dump(perf_stats, f)

def get_data_from_file(idnum, bucket_name='temp_btzoo', directory='results'):
    """
    Get data from google cloud given id number
    idnum
        id of the backtest
    bucket_name
        name of the bucket
    directory
        directory from which files would be fetched
    returns the content of the HDF file
    """
    if not(GOOGLE_CLOUD):
        print('Google Cloud library not installed')
        return
    bucket = client.bucket(bucket_name)
    blob = bucket.blob('{}/{}.h5'.format(directory, idnum))
    fp = tempfile.NamedTemporaryFile()
    blob.download_to_filename(fp.name)
    return pd.read_hdf(fp.name)  



def main():
    return
    benchmark = get_benchmark()
    counter = 0
    for root,directory,files in os.walk(DIR):
        for file in files:
            if file.endswith('h5'):
                filename = os.path.join(root, file)
                output_filename = os.path.join(OUTPUT_DIR, file.split('.')[0])+'.json'
                counter+=1
                # Using thread pool since this is an I/O bound task
                with concurrent.futures.ThreadPoolExecutor() as executor:
                    executor.submit(runner, filename, output_filename, benchmark, counter)

if __name__ == "__main__":
    DIR = '/media/machine/4E1EA2D152455460/temp/btzoo_results/results'
    OUTPUT_DIR = '/media/machine/4E1EA2D152455460/temp/btzoo_results/summary'
    main()