adelorenzo/stable
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

first step toward cleaning output and enabling multimodel training per pair

Browse Source
This commit is contained in:
robcaulk 2022-07-01 14:00:30 +02:00
parent 6c7d02cb18
commit 93e1410ed9
3 changed files with 63 additions and 144 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

73
freqtrade/freqai/data_drawer.py
View File

@ -1,6 +1,5 @@
import collections import collections
import copy
import json import json
import logging import logging
import re import re
@ -11,6 +10,7 @@ from typing import Any, Dict, Tuple
# import pickle as pk # import pickle as pk
import numpy as np import numpy as np
import pandas as pd
from pandas import DataFrame from pandas import DataFrame
@ -163,18 +163,13 @@ class FreqaiDataDrawer:
# send pair to end of queue # send pair to end of queue
self.pair_dict[pair]['priority'] = len(self.pair_dict) self.pair_dict[pair]['priority'] = len(self.pair_dict)
def set_initial_return_values(self, pair: str, dh): def set_initial_return_values(self, pair: str, dh, dataframe: DataFrame) -> None:
self.model_return_values[pair] = {} self.model_return_values[pair] = dataframe
self.model_return_values[pair]['predictions'] = dh.full_predictions self.model_return_values[pair]['target_mean'] = dh.data['target_mean']
self.model_return_values[pair]['do_preds'] = dh.full_do_predict self.model_return_values[pair]['target_std'] = dh.data['target_std']
self.model_return_values[pair]['target_mean'] = dh.full_target_mean
self.model_return_values[pair]['target_std'] = dh.full_target_std
if self.freqai_info.get('feature_parameters', {}).get('DI_threshold', 0) > 0: if self.freqai_info.get('feature_parameters', {}).get('DI_threshold', 0) > 0:
self.model_return_values[pair]['DI_values'] = dh.full_DI_values self.model_return_values[pair]['DI_values'] = dh.DI_values
# if not self.follow_mode:
# self.save_model_return_values_to_disk()
def append_model_predictions(self, pair: str, predictions, do_preds, def append_model_predictions(self, pair: str, predictions, do_preds,
target_mean, target_std, dh, len_df) -> None: target_mean, target_std, dh, len_df) -> None:
@ -182,7 +177,7 @@ class FreqaiDataDrawer:
# strat seems to feed us variable sized dataframes - and since we are trying to build our # strat seems to feed us variable sized dataframes - and since we are trying to build our
# own return array in the same shape, we need to figure out how the size has changed # own return array in the same shape, we need to figure out how the size has changed
# and adapt our stored/returned info accordingly. # and adapt our stored/returned info accordingly.
length_difference = len(self.model_return_values[pair]['predictions']) - len_df length_difference = len(self.model_return_values[pair]['prediction']) - len_df
i = 0 i = 0
if length_difference == 0: if length_difference == 0:
@ -190,51 +185,29 @@ class FreqaiDataDrawer:
elif length_difference > 0: elif length_difference > 0:
i = length_difference + 1 i = length_difference + 1
self.model_return_values[pair]['predictions'] = np.append( df = self.model_return_values[pair].shift(-i)
self.model_return_values[pair]['predictions'][i:], predictions[-1])
df['prediction'].iloc[-1] = predictions[-1]
df['do_predict'].iloc[-1] = do_preds[-1]
df['target_mean'].iloc[-1] = target_mean
df['target_std'].iloc[-1] = target_std
if self.freqai_info.get('feature_parameters', {}).get('DI_threshold', 0) > 0: if self.freqai_info.get('feature_parameters', {}).get('DI_threshold', 0) > 0:
self.model_return_values[pair]['DI_values'] = np.append( df['DI_values'].iloc[-1] = dh.DI_values[-1]
self.model_return_values[pair]['DI_values'][i:], dh.DI_values[-1])
self.model_return_values[pair]['do_preds'] = np.append(
self.model_return_values[pair]['do_preds'][i:], do_preds[-1])
self.model_return_values[pair]['target_mean'] = np.append(
self.model_return_values[pair]['target_mean'][i:], target_mean)
self.model_return_values[pair]['target_std'] = np.append(
self.model_return_values[pair]['target_std'][i:], target_std)
if length_difference < 0: if length_difference < 0:
prepend = np.zeros(abs(length_difference) - 1) prepend_df = pd.DataFrame(np.zeros((abs(length_difference) - 1, len(df.columns))),
self.model_return_values[pair]['predictions'] = np.insert( columns=df.columns)
self.model_return_values[pair]['predictions'], 0, prepend) df = pd.concat([prepend_df, df], axis=0)
if self.freqai_info.get('feature_parameters', {}).get('DI_threshold', 0) > 0:
self.model_return_values[pair]['DI_values'] = np.insert(
self.model_return_values[pair]['DI_values'], 0, prepend)
self.model_return_values[pair]['do_preds'] = np.insert(
self.model_return_values[pair]['do_preds'], 0, prepend)
self.model_return_values[pair]['target_mean'] = np.insert(
self.model_return_values[pair]['target_mean'], 0, prepend)
self.model_return_values[pair]['target_std'] = np.insert(
self.model_return_values[pair]['target_std'], 0, prepend)
dh.full_predictions = copy.deepcopy(self.model_return_values[pair]['predictions'])
dh.full_do_predict = copy.deepcopy(self.model_return_values[pair]['do_preds'])
dh.full_target_mean = copy.deepcopy(self.model_return_values[pair]['target_mean'])
dh.full_target_std = copy.deepcopy(self.model_return_values[pair]['target_std'])
if self.freqai_info.get('feature_parameters', {}).get('DI_threshold', 0) > 0:
dh.full_DI_values = copy.deepcopy(self.model_return_values[pair]['DI_values'])
# if not self.follow_mode:
# self.save_model_return_values_to_disk()
def return_null_values_to_strategy(self, dataframe: DataFrame, dh) -> None: def return_null_values_to_strategy(self, dataframe: DataFrame, dh) -> None:
len_df = len(dataframe) dataframe['prediction'] = 0
dh.full_predictions = np.zeros(len_df) dataframe['do_predict'] = 0
dh.full_do_predict = np.zeros(len_df) dataframe['target_mean'] = 0
dh.full_target_mean = np.zeros(len_df) dataframe['target_std'] = 0
dh.full_target_std = np.zeros(len_df)
if self.freqai_info.get('feature_parameters', {}).get('DI_threshold', 0) > 0: if self.freqai_info.get('feature_parameters', {}).get('DI_threshold', 0) > 0:
dh.full_DI_values = np.zeros(len_df) dataframe['DI_value'] = 0
def purge_old_models(self) -> None: def purge_old_models(self) -> None:

20
freqtrade/freqai/data_kitchen.py
View File

@ -88,7 +88,7 @@ class FreqaiDataKitchen:
return return
def save_data(self, model: Any, coin: str = '') -> None: def save_data(self, model: Any, coin: str = '', keras_model=False) -> None:
""" """
Saves all data associated with a model for a single sub-train time range Saves all data associated with a model for a single sub-train time range
:params: :params:
@ -102,7 +102,10 @@ class FreqaiDataKitchen:
save_path = Path(self.data_path) save_path = Path(self.data_path)
# Save the trained model # Save the trained model
dump(model, save_path / str(self.model_filename + "_model.joblib")) if not keras_model:
dump(model, save_path / str(self.model_filename + "_model.joblib"))
else:
model.save(save_path / str(self.model_filename + "_model.h5"))
if self.svm_model is not None: if self.svm_model is not None:
dump(self.svm_model, save_path / str(self.model_filename + "_svm_model.joblib")) dump(self.svm_model, save_path / str(self.model_filename + "_svm_model.joblib"))
@ -144,7 +147,7 @@ class FreqaiDataKitchen:
return return
def load_data(self, coin: str = '') -> Any: def load_data(self, coin: str = '', keras_model=False) -> Any:
""" """
loads all data required to make a prediction on a sub-train time range loads all data required to make a prediction on a sub-train time range
:returns: :returns:
@ -190,8 +193,11 @@ class FreqaiDataKitchen:
# try to access model in memory instead of loading object from disk to save time # try to access model in memory instead of loading object from disk to save time
if self.live and self.model_filename in self.data_drawer.model_dictionary: if self.live and self.model_filename in self.data_drawer.model_dictionary:
model = self.data_drawer.model_dictionary[self.model_filename] model = self.data_drawer.model_dictionary[self.model_filename]
else: elif not keras_model:
model = load(self.data_path / str(self.model_filename + "_model.joblib")) model = load(self.data_path / str(self.model_filename + "_model.joblib"))
else:
from tensorflow import keras
model = keras.models.load_model(self.data_path / str(self.model_filename + "_model.h5"))
if Path(self.data_path / str(self.model_filename + if Path(self.data_path / str(self.model_filename +
"_svm_model.joblib")).resolve().exists(): "_svm_model.joblib")).resolve().exists():
@ -287,7 +293,11 @@ class FreqaiDataKitchen:
training_filter training_filter
): # we don't care about total row number (total no. datapoints) in training, we only care ): # we don't care about total row number (total no. datapoints) in training, we only care
# about removing any row with NaNs # about removing any row with NaNs
drop_index_labels = pd.isnull(labels) # if labels has multiple columns (user wants to train multiple models), we detect here
if labels.shape[1] == 1:
drop_index_labels = pd.isnull(labels)
else:
drop_index_labels = pd.isnull(labels).any(1)
drop_index_labels = drop_index_labels.replace(True, 1).replace(False, 0) drop_index_labels = drop_index_labels.replace(True, 1).replace(False, 0)
filtered_dataframe = filtered_dataframe[ filtered_dataframe = filtered_dataframe[
(drop_index == 0) & (drop_index_labels == 0) (drop_index == 0) & (drop_index_labels == 0)

114
freqtrade/freqai/freqai_interface.py
View File

@ -68,6 +68,8 @@ class IFreqaiModel(ABC):
self.scanning = False self.scanning = False
self.ready_to_scan = False self.ready_to_scan = False
self.first = True self.first = True
self.keras = self.freqai_info.get('keras', False)
self.CONV_WIDTH = self.freqai_info.get('conv_width', 2)
def assert_config(self, config: Dict[str, Any]) -> None: def assert_config(self, config: Dict[str, Any]) -> None:
@ -122,30 +124,28 @@ class IFreqaiModel(ABC):
time.sleep(1) time.sleep(1)
for pair in self.config.get('exchange', {}).get('pair_whitelist'): for pair in self.config.get('exchange', {}).get('pair_whitelist'):
(model_filename, (_, trained_timestamp, _, _) = self.data_drawer.get_pair_dict_info(pair)
trained_timestamp,
_, _) = self.data_drawer.get_pair_dict_info(pair)
if self.data_drawer.pair_dict[pair]['priority'] != 1: if self.data_drawer.pair_dict[pair]['priority'] != 1:
continue continue
dh = FreqaiDataKitchen(self.config, self.data_drawer, dh = FreqaiDataKitchen(self.config, self.data_drawer,
self.live, pair) self.live, pair)
file_exists = False # file_exists = False
dh.set_paths(pair, trained_timestamp) dh.set_paths(pair, trained_timestamp)
file_exists = self.model_exists(pair, # file_exists = self.model_exists(pair,
dh, # dh,
trained_timestamp=trained_timestamp, # trained_timestamp=trained_timestamp,
model_filename=model_filename, # model_filename=model_filename,
scanning=True) # scanning=True)
(retrain, (retrain,
new_trained_timerange, new_trained_timerange,
data_load_timerange) = dh.check_if_new_training_required(trained_timestamp) data_load_timerange) = dh.check_if_new_training_required(trained_timestamp)
dh.set_paths(pair, new_trained_timerange.stopts) dh.set_paths(pair, new_trained_timerange.stopts)
if retrain or not file_exists: if retrain: # or not file_exists:
self.train_model_in_series(new_trained_timerange, pair, self.train_model_in_series(new_trained_timerange, pair,
strategy, dh, data_load_timerange) strategy, dh, data_load_timerange)
@ -199,9 +199,9 @@ class IFreqaiModel(ABC):
self.data_drawer.pair_dict[metadata['pair']][ self.data_drawer.pair_dict[metadata['pair']][
'trained_timestamp'] = trained_timestamp.stopts 'trained_timestamp'] = trained_timestamp.stopts
dh.set_new_model_names(metadata['pair'], trained_timestamp) dh.set_new_model_names(metadata['pair'], trained_timestamp)
dh.save_data(self.model, metadata['pair']) dh.save_data(self.model, metadata['pair'], keras=self.keras)
else: else:
self.model = dh.load_data(metadata['pair']) self.model = dh.load_data(metadata['pair'], keras=self.keras)
self.check_if_feature_list_matches_strategy(dataframe_train, dh) self.check_if_feature_list_matches_strategy(dataframe_train, dh)
@ -278,7 +278,7 @@ class IFreqaiModel(ABC):
f'using { self.identifier }') f'using { self.identifier }')
# load the model and associated data into the data kitchen # load the model and associated data into the data kitchen
self.model = dh.load_data(coin=metadata['pair']) self.model = dh.load_data(coin=metadata['pair'], keras=self.keras)
if not self.model: if not self.model:
logger.warning('No model ready, returning null values to strategy.') logger.warning('No model ready, returning null values to strategy.')
@ -297,14 +297,16 @@ class IFreqaiModel(ABC):
trained_timestamp: int) -> None: trained_timestamp: int) -> None:
# hold the historical predictions in memory so we are sending back # hold the historical predictions in memory so we are sending back
# correct array to strategy FIXME currently broken, but only affecting # correct array to strategy
# Frequi reporting. Signals remain unaffeted.
if pair not in self.data_drawer.model_return_values: if pair not in self.data_drawer.model_return_values:
preds, do_preds = self.predict(dataframe, dh) preds, do_preds = self.predict(dataframe, dh)
dh.append_predictions(preds, do_preds, len(dataframe)) # mypy doesnt like the typing in else statement, so we need to explicitly add to
dh.fill_predictions(len(dataframe)) # dataframe separately
self.data_drawer.set_initial_return_values(pair, dh) dataframe['prediction'], dataframe['do_predict'] = preds, do_preds
# dh.append_predictions(preds, do_preds, len(dataframe))
# dh.fill_predictions(len(dataframe))
self.data_drawer.set_initial_return_values(pair, dh, dataframe)
return return
elif self.dh.check_if_model_expired(trained_timestamp): elif self.dh.check_if_model_expired(trained_timestamp):
preds, do_preds, dh.DI_values = np.zeros(2), np.ones(2) * 2, np.zeros(2) preds, do_preds, dh.DI_values = np.zeros(2), np.ones(2) * 2, np.zeros(2)
@ -312,7 +314,8 @@ class IFreqaiModel(ABC):
'construction should take care to consider this event with ' 'construction should take care to consider this event with '
'prediction == 0 and do_predict == 2') 'prediction == 0 and do_predict == 2')
else: else:
preds, do_preds = self.predict(dataframe.iloc[-2:], dh) # Only feed in the most recent candle for prediction in live scenario
preds, do_preds = self.predict(dataframe.iloc[-self.CONV_WIDTH:], dh, first=False)
self.data_drawer.append_model_predictions(pair, preds, do_preds, self.data_drawer.append_model_predictions(pair, preds, do_preds,
dh.data["target_mean"], dh.data["target_mean"],
@ -426,71 +429,6 @@ class IFreqaiModel(ABC):
if not col.startswith('%') or col.startswith('%%')] if not col.startswith('%') or col.startswith('%%')]
return dataframe[to_keep] return dataframe[to_keep]
@threaded
def retrain_model_on_separate_thread(self, new_trained_timerange: TimeRange, pair: str,
strategy: IStrategy, dh: FreqaiDataKitchen,
data_load_timerange: TimeRange):
"""
Retreive data and train model on separate thread. Always called if the model folder already
contains a full set of trained models.
:params:
new_trained_timerange: TimeRange = the timerange to train the model on
metadata: dict = strategy provided metadata
strategy: IStrategy = user defined strategy object
dh: FreqaiDataKitchen = non-persistent data container for current coin/loop
data_load_timerange: TimeRange = the amount of data to be loaded for populate_any_indicators
(larger than new_trained_timerange so that new_trained_timerange does not contain any NaNs)
"""
# with nostdout():
# dh.download_new_data_for_retraining(data_load_timerange, metadata, strategy)
# corr_dataframes, base_dataframes = dh.load_pairs_histories(data_load_timerange,
# metadata)
corr_dataframes, base_dataframes = dh.get_base_and_corr_dataframes(data_load_timerange,
pair)
# protecting from common benign errors associated with grabbing new data from exchange:
try:
unfiltered_dataframe = dh.use_strategy_to_populate_indicators(strategy,
corr_dataframes,
base_dataframes,
pair)
unfiltered_dataframe = dh.slice_dataframe(new_trained_timerange, unfiltered_dataframe)
except Exception as err:
logger.exception(err)
self.training_on_separate_thread = False
self.retrain = False
return
try:
model = self.train(unfiltered_dataframe, pair, dh)
except ValueError:
logger.warning('Value error encountered during training')
self.training_on_separate_thread = False
self.retrain = False
return
self.data_drawer.pair_dict[pair][
'trained_timestamp'] = new_trained_timerange.stopts
dh.set_new_model_names(pair, new_trained_timerange)
# logger.info('Training queue'
# f'{sorted(self.data_drawer.pair_dict.items(), key=lambda item: item[1])}')
if self.data_drawer.pair_dict[pair]['priority'] == 1:
with self.lock:
self.data_drawer.pair_to_end_of_training_queue(pair)
dh.save_data(model, coin=pair)
# self.training_on_separate_thread = False
# self.retrain = False
# each time we finish a training, we check the directory to purge old models.
if self.freqai_info.get('purge_old_models', False):
self.data_drawer.purge_old_models()
return
def train_model_in_series(self, new_trained_timerange: TimeRange, pair: str, def train_model_in_series(self, new_trained_timerange: TimeRange, pair: str,
strategy: IStrategy, dh: FreqaiDataKitchen, strategy: IStrategy, dh: FreqaiDataKitchen,
data_load_timerange: TimeRange): data_load_timerange: TimeRange):
@ -505,9 +443,7 @@ class IFreqaiModel(ABC):
data_load_timerange: TimeRange = the amount of data to be loaded for populate_any_indicators data_load_timerange: TimeRange = the amount of data to be loaded for populate_any_indicators
(larger than new_trained_timerange so that new_trained_timerange does not contain any NaNs) (larger than new_trained_timerange so that new_trained_timerange does not contain any NaNs)
""" """
# dh.download_new_data_for_retraining(data_load_timerange, metadata, strategy)
# corr_dataframes, base_dataframes = dh.load_pairs_histories(data_load_timerange,
# metadata)
corr_dataframes, base_dataframes = dh.get_base_and_corr_dataframes(data_load_timerange, corr_dataframes, base_dataframes = dh.get_base_and_corr_dataframes(data_load_timerange,
pair) pair)
@ -527,7 +463,7 @@ class IFreqaiModel(ABC):
if self.data_drawer.pair_dict[pair]['priority'] == 1 and self.scanning: if self.data_drawer.pair_dict[pair]['priority'] == 1 and self.scanning:
with self.lock: with self.lock:
self.data_drawer.pair_to_end_of_training_queue(pair) self.data_drawer.pair_to_end_of_training_queue(pair)
dh.save_data(model, coin=pair) dh.save_data(model, coin=pair, keras=self.keras)
if self.freqai_info.get('purge_old_models', False): if self.freqai_info.get('purge_old_models', False):
self.data_drawer.purge_old_models() self.data_drawer.purge_old_models()
@ -563,7 +499,7 @@ class IFreqaiModel(ABC):
@abstractmethod @abstractmethod
def predict(self, dataframe: DataFrame, def predict(self, dataframe: DataFrame,
dh: FreqaiDataKitchen) -> Tuple[npt.ArrayLike, npt.ArrayLike]: dh: FreqaiDataKitchen, first: bool = True) -> Tuple[npt.ArrayLike, npt.ArrayLike]:
""" """
Filter the prediction features data and predict with it. Filter the prediction features data and predict with it.
:param: :param: