Python GruSettings.batch_size Examples

Example #1

File: lstm_final_lb.py Project: JakobStruye/timeseries-comparison

import sys
import shared

sys.path.append(".") #..welp
from run_rnn import  GruSettings, run_gru


settings = GruSettings()
settings.max_verbosity = 2
settings.epochs = 200
settings.epochs_retrain = 60
settings.online = True
nodes = 200
lr = 0.003#231262326395
settings.nodes = nodes
settings.batch_size = 128
settings.lr = lr
settings.loss = 'mae'
settings.stateful=False
settings.lookback = 50
settings.lookback_as_features=False
settings.feature_count = 3
settings.predictionStep = 5
settings.season = 48
settings.adam_eps = 0.001
settings.retrain_interval = 1000
settings.reset_on_retrain = False
settings.refeed_on_retrain = True
settings.cutoff_normalize = True
settings.use_dropout = True
settings.nTrain = 5000

Example #2

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'  #Suppress build warnings
import sys
import shared

sys.path.append(".")  #..welp
from run_rnn import GruSettings, run_gru

settings = GruSettings()
settings.max_verbosity = 2
settings.epochs = 400
settings.epochs_retrain = 125
settings.online = True
nodes = 180
lr = 0.0015
settings.nodes = nodes
settings.batch_size = 512
settings.lr = lr
settings.loss = 'mae'
settings.stateful = False
settings.lookback = 75
settings.lookback_as_features = True
settings.feature_count = 3
settings.predictionStep = 5
settings.season = 48
settings.adam_eps = 0.001
settings.retrain_interval = 1000
settings.reset_on_retrain = False
settings.refeed_on_retrain = True
settings.cutoff_normalize = True
settings.use_dropout = True
settings.nTrain = 5000

Example #3

File: lstm_final_lb_ft_quick.py Project: JakobStruye/timeseries-comparison

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'  #Suppress build warnings
import sys
import shared

sys.path.append(".")  #..welp
from run_rnn import GruSettings, run_gru

settings = GruSettings()
settings.max_verbosity = 2
settings.epochs = 75
settings.online = True
nodes = 50
lr = 0.02
settings.nodes = nodes
settings.batch_size = 256
settings.lr = lr
settings.loss = 'mae'
settings.stateful = False
settings.lookback = 75
settings.lookback_as_features = True
settings.feature_count = 3
settings.predictionStep = 5
settings.season = 48
settings.adam_eps = 0.001
settings.retrain_interval = 2500
settings.reset_on_retrain = True  #TEMP MODIFIED
settings.refeed_on_retrain = True
settings.cutoff_normalize = True
settings.nTrain = 5000
settings.ignore_for_error = [5500, 10000]

Example #4

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'  #Suppress build warnings
from run_rnn import GruSettings, run_gru

settings = GruSettings()
settings.max_verbosity = 2
settings.epochs = 50
settings.epochs_retrain = 25
settings.online = True
nodes = 40  #317
lr = 0.003  #0.0031802801373# 0.0077
settings.nodes = nodes
settings.batch_size = 1024
settings.lr = lr
settings.loss = 'mae'
settings.stateful = False
settings.lookback = 60  #None to not use lookbacks
settings.lookback_as_features = True  #Considers each lookback value a separate feature if True, ignored if lookback is None
settings.feature_count = 3  #Uses the first `feature_count` selected columns as features, ignored if `lookback_as_features`
settings.predictionStep = 5
settings.season = 48
settings.adam_eps = 0.001
settings.retrain_interval = 1000
settings.reset_on_retrain = False
settings.onego_refeed_on_retrain = True
settings.cutoff_normalize = True
settings.use_dropout = False
settings.nTrain = 5000
settings.ignore_for_error = [5500, 10000]
settings.normalization_type = 'default'
settings.implementation = 'keras'

Example #5

File: run_gru_mase.py Project: JakobStruye/timeseries-comparison

10: lb_as_features
11: feature_count
12: implementation 
13: adam epsilon
"""
settings = GruSettings()
settings.max_verbosity = 0
settings.epochs = int(argv[7])
settings.online = True if argv[8] == "True" else False
try:
    settings.lookback = int(argv[6])
except:
    settings.lookback = None
settings.nodes = int(argv[3])
try:
    settings.batch_size = int(argv[9])
except:
    settings.batch_size = None
settings.retrain_interval = int(argv[4])
settings.lookback_as_features = True if argv[10] == "True" else False
settings.feature_count = int(argv[11])
settings.lr = float(argv[5])
settings.predictionStep = 5
settings.season = 48
settings.ignore_for_error = [5500]
settings.nTrain = 5000
settings.limit_to = 7500  # if not
settings.normalization_type = 'default'
settings.implementation = 'keras'
settings.rnn_type = argv[12]
settings.use_binary = False