diff --git a/catalyst/curate/poloniex.py b/catalyst/curate/poloniex.py index e92a9a11..911ac25f 100644 --- a/catalyst/curate/poloniex.py +++ b/catalyst/curate/poloniex.py @@ -1,12 +1,10 @@ import json, time, csv from datetime import datetime import pandas as pd -import os -import time -import requests -import logbook +import os, time, shutil, requests, logbook -DT_START = time.mktime(datetime(2010, 1, 1, 0, 0).timetuple()) +DT_START = int(time.mktime(datetime(2010, 1, 1, 0, 0).timetuple())) +DT_END = int(time.time()) CSV_OUT_FOLDER = '/var/tmp/catalyst/data/poloniex/' CONN_RETRIES = 2 @@ -14,9 +12,9 @@ logbook.StderrHandler().push_application() log = logbook.Logger(__name__) class PoloniexCurator(object): - """ + ''' OHLCV data feed generator for crypto data. Based on Poloniex market data - """ + ''' _api_path = 'https://poloniex.com/public?' currency_pairs = [] @@ -29,6 +27,9 @@ class PoloniexCurator(object): log.error('Failed to create data folder: %s' % CSV_OUT_FOLDER) log.exception(e) + ''' + Retrieves and returns all currency pairs from the exchange + ''' def get_currency_pairs(self): url = self._api_path + 'command=returnTicker' @@ -47,98 +48,210 @@ class PoloniexCurator(object): log.debug('Currency pairs retrieved successfully: %d' % (len(self.currency_pairs))) - def _get_start_date(self, csv_fn): - ''' Function returns latest appended date, if the file has been previously written - the last line is an empty one, so we have to read the second to last line + + ''' + Helper function that reads tradeID and date fields from CSV readline + ''' + def _retrieve_tradeID_date(self, row): + tId = int(row.split(',')[0]) + d = pd.to_datetime( row.split(',')[1], infer_datetime_format=True).value // 10 ** 9 + return tId, d + + ''' + Retrieves TradeHistory from exchange for a given currencyPair between start and end dates. + If no start date is provided, uses a system-wide one (beginning of time for cryptotrading) + If no end date is provided, 'now' is used + Stores results in CSV file on disk. + This function is called recursively to work around the limitations imposed by the provider API. + ''' + def retrieve_trade_history(self, currencyPair, start=DT_START, end=DT_END, temp=None): + csv_fn = CSV_OUT_FOLDER + 'crypto_trades-' + currencyPair + '.csv' + + ''' + Check what data we already have on disk, reading first and last lines from file. + Data is stored on file from NEWEST to OLDEST. ''' try: with open(csv_fn, 'ab+') as f: - f.seek(0, os.SEEK_END) # First check file is not zero size - if(f.tell() > 2): - f.seek(-2, os.SEEK_END) # Jump to the second last byte. - while f.read(1) != b"\n": # Until EOL is found... - f.seek(-2, os.SEEK_CUR) # ...jump back the read byte plus one more. - lastrow = f.readline() - return int(lastrow.split(',')[0]) + 300 + f.seek(0, os.SEEK_END) + if(f.tell() > 2): # First check file is not zero size + f.seek(0) # Go to the beginning to read first line + last_tradeID, end_file = self._retrieve_tradeID_date(f.readline()) + f.seek(-2, os.SEEK_END) # Jump to the second last byte. + while f.read(1) != b"\n": # Until EOL is found... + f.seek(-2, os.SEEK_CUR) # ...jump back the read byte plus one more. + first_tradeID, start_file = self._retrieve_tradeID_date(f.readline()) + + if( first_tradeID == 1 and end_file + 3600 > DT_END ): + return except Exception as e: log.error('Error opening file: %s' % csv_fn) log.exception(e) - return DT_START + ''' + Poloniex API limits querying TradeHistory to intervals smaller than 1 month, + so we make sure that start date is never more than 1 month apart from end date + ''' + if( end - start > 2419200 ): # 60 s/min * 60 min/hr * 24 hr/day * 28 days + newstart = end - 2419200 + else: + newstart = start - def get_data(self, currencyPair, start, end=9999999999, period=300): - url = self._api_path + 'command=returnChartData¤cyPair=' + currencyPair + '&start=' + str(start) + '&end=' + str(end) + '&period=' + str(period) + log.debug(currencyPair+': Retrieving from '+str(newstart)+' to '+str(end) +'\t ' + + time.ctime(newstart) + ' - '+ time.ctime(end)) + + url = self._api_path + 'command=returnTradeHistory¤cyPair=' + currencyPair + '&start=' + str(newstart) + '&end=' + str(end) try: response = requests.get(url) except Exception as e: - log.error('Failed to retrieve candlestick chart data for %s' % currencyPair) + log.error('Failed to retrieve trade history data for %s' % currencyPair) log.exception(e) return None + else: + if isinstance(response.json(), dict) and response.json()['error']: + log.error('Failed to to retrieve trade history data for %s: %s' % (currencyPair,response.json()['error'])) + exit(1) + + ''' + If we get to transactionId == 1, and we already have that on disk, + we got to the end of TradeHistory for this coin. + ''' + if('first_tradeID' in locals() and response.json()[-1]['tradeID'] == first_tradeID): + return + + ''' + There are primarily two scenarios: + a) There is newer data available that we need to add at the beginning + of the file. We'll retrieve all what we need until we get to what + we already have, writing it to a temporary file; and we will write + that at the beginning of our existing file. + b) We are going back in time, appending at the end of our existing + TradeHistory until the first transaction for this currencyPair + ''' + try: + if( 'end_file' in locals() and end_file + 3600 < end): + if (temp is None): + temp = os.tmpfile() + tempcsv = csv.writer(temp) + for item in response.json(): + if( item['tradeID'] <= last_tradeID ): + continue + tempcsv.writerow([ + item['tradeID'], + item['date'], + item['type'], + item['rate'], + item['amount'], + item['total'], + item['globalTradeID'] + ]) + if( response.json()[-1]['tradeID'] > last_tradeID ): + end = pd.to_datetime( response.json()[-1]['date'], infer_datetime_format=True).value // 10 ** 9 + self.retrieve_trade_history(currencyPair, start, end, temp=temp) + else: + with open(csv_fn,'rb+') as f: + shutil.copyfileobj(f,temp) + f.seek(0) + temp.seek(0) + shutil.copyfileobj(temp,f) + temp.close() + end = start_file + else: + with open(csv_fn, 'ab') as csvfile: + csvwriter = csv.writer(csvfile) + for item in response.json(): + if( 'first_tradeID' in locals() and item['tradeID'] >= first_tradeID ): + continue + csvwriter.writerow([ + item['tradeID'], + item['date'], + item['type'], + item['rate'], + item['amount'], + item['total'], + item['globalTradeID'] + ]) + end = pd.to_datetime( response.json()[-1]['date'], infer_datetime_format=True).value // 10 ** 9 + + except Exception as e: + log.error('Error opening %s' % csv_fn) + log.exception(e) + + ''' + If we got here, we aren't done yet. Call recursively with 'end' times + that go sequentially back in time. + ''' + self.retrieve_trade_history(currencyPair, start, end) - return response.json() ''' - Pulls latest data for a single pair + Generates OHLCV dataframe from a dataframe containing all TradeHistory + by resampling with 1-minute period ''' - def append_data_single_pair(self, currencyPair, repeat=0): - log.debug('Getting data for %s' % currencyPair) - csv_fn = CSV_OUT_FOLDER + 'crypto_prices-' + currencyPair + '.csv' - start = self._get_start_date(csv_fn) - # Only fetch data if more than 5min have passed since last fetch - if (time.time() > start): - data = self.get_data(currencyPair, start) - if data is not None: - try: - with open(csv_fn, 'ab') as csvfile: - csvwriter = csv.writer(csvfile) - for item in data: - if item['date'] == 0: - continue - csvwriter.writerow([ - item['date'], - item['open'], - item['high'], - item['low'], - item['close'], - item['volume'], - ]) - except Exception as e: - log.error('Error opening %s' % csv_fn) - log.exception(e) - elif (repeat < CONN_RETRIES): - log.debug('Retrying: attemt %d' % (repeat+1) ) - self.append_data_single_pair(currencyPair, repeat + 1) + def generate_ohlcv(self, df): + df.set_index('date', inplace=True) # Index by date + vol = df['total'].to_frame('volume') # Will deal with vol separately, as ohlc() messes it up + df.drop('total', axis=1, inplace=True) # Drop volume data from dataframe + ohlc = df.resample('T').ohlc() # Resample OHLC in 1min bins + ohlc.columns = ohlc.columns.map(lambda t: t[1]) # Raname columns by dropping 'rate' + closes = ohlc['close'].fillna(method='pad') # Pad forward missing 'close' + ohlc = ohlc.apply(lambda x: x.fillna(closes)) # Fill N/A with last close + vol = vol.resample('T').sum().fillna(0) # Add volumes by bin + ohlcv = pd.concat([ohlc,vol], axis=1) # Concatenate OHLC + Volume + return ohlcv + ''' - Pulls latest data for all currency pairs + Generates OHLCV data file with 1minute bars from TradeHistory on disk ''' - def append_data(self): - for currencyPair in self.currency_pairs: - self.append_data_single_pair(currencyPair) - # Rate limit is 6 calls per second, sleep 1sec/6 to be safe - time.sleep(0.17) + def write_ohlcv_file(self, currencyPair): + csv_trades = CSV_OUT_FOLDER + 'crypto_trades-' + currencyPair + '.csv' + csv_1min = CSV_OUT_FOLDER + 'crypto_1min-' + currencyPair + '.csv' + if( os.path.isfile(csv_1min) ): + log.debug(currencyPair+': 1min data already present. Delete the file if you want to rebuild it.') + else: + df = pd.read_csv(csv_trades, names=['tradeID','date','type','rate','amount','total','globalTradeID'], + dtype = {'tradeID': int, 'date': str, 'type': str, 'rate': float, 'amount': float, 'total': float, 'globalTradeID': int } ) + df.drop(['tradeID','type','amount','globalTradeID'], axis=1, inplace=True) + df['date'] = pd.to_datetime(df['date'], infer_datetime_format=True) + ohlcv = self.generate_ohlcv(df) + try: + with open(csv_1min, 'ab') as csvfile: + csvwriter = csv.writer(csvfile) + for item in ohlcv.itertuples(): + if item.Index == 0: + continue + csvwriter.writerow([ + item.Index.value // 10 ** 9, + item.open, + item.high, + item.low, + item.close, + item.volume, + ]) + except Exception as e: + log.error('Error opening %s' % csv_fn) + log.exception(e) + log.debug(currencyPair+': Generated 1min OHLCV data.') + ''' - Returns a data frame for all pairs, or for the requests currency pair. - Makes sure data is up to date + Returns a data frame for a given currencyPair from data on disk ''' - def to_dataframe(self, start, end, currencyPair=None): - csv_fn = CSV_OUT_FOLDER + 'crypto_prices-' + currencyPair + '.csv' - last_date = self._get_start_date(csv_fn) - if last_date + 300 < end or not os.path.exists(csv_fn): - # get latest data - self.append_data_single_pair(currencyPair) - - # CSV holds the latest snapshot - df = pd.read_csv(csv_fn, names=['date', 'open', 'high', 'low', 'close', 'volume']) - df['date']=pd.to_datetime(df['date'],unit='s') + def onemin_to_dataframe(self, currencyPair, start, end): + csv_fn = CSV_OUT_FOLDER + 'crypto_1min-' + currencyPair + '.csv' + df = pd.read_csv(csv_fn, names=['date', 'open', 'high', 'low', 'close', 'volume']) + df['date'] = pd.to_datetime(df['date'],unit='s') df.set_index('date', inplace=True) + return df[start : end] - return df[datetime.fromtimestamp(start):datetime.fromtimestamp(end-1)] if __name__ == '__main__': pc = PoloniexCurator() pc.get_currency_pairs() - pc.append_data() + + for currencyPair in pc.currency_pairs: + pc.retrieve_trade_history(currencyPair) + pc.write_ohlcv_file(currencyPair) diff --git a/catalyst/data/bundles/poloniex.py b/catalyst/data/bundles/poloniex.py index 9ad60d01..60bbac95 100644 --- a/catalyst/data/bundles/poloniex.py +++ b/catalyst/data/bundles/poloniex.py @@ -25,6 +25,8 @@ from catalyst.data.bundles.core import register_bundle from catalyst.data.bundles.base_pricing import BaseCryptoPricingBundle from catalyst.utils.memoize import lazyval +from catalyst.curate.poloniex import PoloniexCurator + class PoloniexBundle(BaseCryptoPricingBundle): @lazyval def name(self): @@ -38,7 +40,7 @@ class PoloniexBundle(BaseCryptoPricingBundle): def frequencies(self): return set(( 'daily', - #'5-minute', + 'minute', )) @lazyval @@ -94,17 +96,23 @@ class PoloniexBundle(BaseCryptoPricingBundle): start_date, end_date, frequency): - raw = pd.read_json( - self._format_data_url( - api_key, - symbol, - start_date, - end_date, - frequency, - ), - orient='records', - ) - raw.set_index('date', inplace=True) + + if(frequency == 'minute'): + pc = PoloniexCurator() + raw = pc.onemin_to_dataframe(symbol, start_date, end_date) + + else: + raw = pd.read_json( + self._format_data_url( + api_key, + symbol, + start_date, + end_date, + frequency, + ), + orient='records', + ) + raw.set_index('date', inplace=True) # BcolzDailyBarReader introduces a 1/1000 factor in the way pricing is stored # on disk, which we compensate here to get the right pricing amounts