From a1bc174740bc629aeb43a984583cea09771e2615 Mon Sep 17 00:00:00 2001 From: Victor Grau Serrat Date: Wed, 20 Sep 2017 15:11:18 -0600 Subject: [PATCH] Wrapping up 1min data for Poloniex in backtesting --- catalyst/curate/poloniex.py | 218 +++++++++--------------------- catalyst/data/bundles/poloniex.py | 32 +++-- 2 files changed, 82 insertions(+), 168 deletions(-) diff --git a/catalyst/curate/poloniex.py b/catalyst/curate/poloniex.py index d63be3f9..911ac25f 100644 --- a/catalyst/curate/poloniex.py +++ b/catalyst/curate/poloniex.py @@ -6,19 +6,15 @@ import os, time, shutil, requests, logbook 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/' -CSV_OUT_FOLDER = '/Volumes/enigma/data/poloniex/' CONN_RETRIES = 2 -COINS = ['USDT_BTC','USDT_DASH','USDT_ETC','USDT_ETH','USDT_LTC','USDT_NXT','USDT_REP','USDT_STR','USDT_XMR','USDT_XRP','USDT_ZEC'] -COINS = ['USDT_BTC',] - 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 = [] @@ -31,6 +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' @@ -49,82 +48,29 @@ 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 - ''' - 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 - - except Exception as e: - log.error('Error opening file: %s' % csv_fn) - log.exception(e) - - return DT_START - - def get_data(self, currencyPair, start, end=9999999999, prev_df=None): - log.debug(currencyPair+': Retrieving from '+str(start)+' to '+str(end)) - - ''' - Poloniex limits a single query to returnTradeHistory to less than a year between start and end - ''' - if(end == 9999999999 and time.time() - start > 365*86400 ): - newstart = time.time() - 360*86400 - elif( end != 9999999999 and end - start > 365*86400 ): - newstart = end - 360*86400 - else: - newstart = start - - 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 trade history data for %s' % currencyPair) - log.exception(e) - return None - - log.debug(currencyPair+': Received '+str(len(response.json()))+' trades.') - if(len(response.json())==1 and not isinstance(response.json(),list)): - r = response.json() - print(r) - if(r['error']): - log.error(r['error']) - return None - - df = pd.DataFrame(data=response.json(), columns = ['date','rate', 'total', 'tradeID']) - df['rate'] = pd.to_numeric( df['rate'], errors='coerce') # Convert rate to float - df['total'] = pd.to_numeric( df['total'], errors='coerce') # Convert vol to float - df['tradeID'] = pd.to_numeric( df['tradeID'], errors='coerce') # Convert vol to float - df['date'] = pd.to_datetime(df['date'], infer_datetime_format=True) # Convert date - df.set_index('tradeID', inplace=True) # Index by tradeID - df = df.iloc[::-1] # Reverse timeseries as TradeHistory is provided newest to oldest - - if(prev_df is not None): - if(prev_df.index[0] == df.index[0]): - return prev_df - df = prev_df.combine_first(df) - - first = df['date'].iloc[0].value // 10 ** 9 - df = self.get_data( currencyPair, start, first, df ) - return df + ''' + 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) @@ -168,9 +114,22 @@ class PoloniexCurator(object): log.error('Failed to to retrieve trade history data for %s: %s' % (currencyPair,response.json()['error'])) exit(1) - if('first_tradeID' in locals() and response.json()[-1]['tradeID'] == first_tradeID): # Got to the end of TradingHistory for this coin + ''' + 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): @@ -220,10 +179,34 @@ class PoloniexCurator(object): log.error('Error opening %s' % csv_fn) log.exception(e) - self.retrieve_trade_history(currencyPair, start, end) # If we get here, we aren't done. Repeat + ''' + 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) - def write_ohlcv_file(self, currencyPair): - + + ''' + Generates OHLCV dataframe from a dataframe containing all TradeHistory + by resampling with 1-minute period + ''' + 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 + + + ''' + Generates OHLCV data file with 1minute bars from TradeHistory on disk + ''' + 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) ): @@ -253,82 +236,10 @@ class PoloniexCurator(object): log.exception(e) log.debug(currencyPair+': Generated 1min OHLCV data.') - 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 5min 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 a single pair - ''' - 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: - ohlcv = self.generate_ohlcv(data) - - try: - with open(csv_fn, '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) - elif (repeat < CONN_RETRIES): - log.debug('Retrying: attemt %d' % (repeat+1) ) - self.append_data_single_pair(currencyPair, repeat + 1) ''' - Pulls latest data for all currency pairs + Returns a data frame for a given currencyPair from data 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) - - ''' - Returns a data frame for all pairs, or for the requested currency pair. - Makes sure data is up to date - ''' - 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') - df.set_index('date', inplace=True) - - return df[datetime.fromtimestamp(start):datetime.fromtimestamp(end-1)] - 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']) @@ -336,16 +247,11 @@ class PoloniexCurator(object): df.set_index('date', inplace=True) return df[start : end] + if __name__ == '__main__': pc = PoloniexCurator() pc.get_currency_pairs() - #pc.append_data() - #for coin in COINS: for currencyPair in pc.currency_pairs: - #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: 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