import logging
import os
import tensorflow as tf
import seaborn as sns
sns.set()
ROOT_FOLDER = configs.ROOT_FOLDER
MODEL_ROOT = configs.MODEL_ROOT
expt_name = "treatment_effects"
# EDIT ME ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Specify parameters for model to load - optimal set from paper listed
action_inputs_only = configs.load_optimal_parameters(
'treatment_rnn_action_inputs_only', expt_name, add_net_name=True)
action_w_trajectory_inputs = configs.load_optimal_parameters('treatment_rnn',
expt_name,
add_net_name=True)
censor_w_action_inputs_only = configs.load_optimal_parameters(
'censor_rnn_action_inputs_only', expt_name, add_net_name=True)
censor_w_trajectory_inputs = configs.load_optimal_parameters('censor_rnn',
expt_name,
add_net_name=True)
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
if __name__ == "__main__":
logging.basicConfig(format='%(levelname)s:%(message)s', level=logging.INFO)
ROOT_FOLDER = configs.ROOT_FOLDER
MODEL_ROOT = configs.MODEL_ROOT
RESULTS_FOLDER = configs.RESULTS_FOLDER
expt_name = "treatment_effects"
logging.basicConfig(format='%(levelname)s:%(message)s', level=logging.INFO)
# EDIT ME! ######################################################################################
# Defines specific parameters to train for - otherwise runs full hyperparameter optimisation
decoder_specifications = {}
# Optimal encoder to load for decoder training
# - This allows for states from the encoder to be dumped, and decoder is intialised with them
encoder_specifications = {
'rnn_propensity_weighted':
configs.load_optimal_parameters('rnn_propensity_weighted', expt_name)
}
# Specify which networks to train - only use R-MSN by default. Full list in activation map
net_names = ['rnn_propensity_weighted']
##################################################################################################
# Data processing Functions ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
def process_seq_data(
data_map,
states,
projection_horizon=5,
num_features_to_include=1e6): # forecast 10 years into the future
def _check_shapes(array1, array2, dims):
import seaborn as sns
sns.set()
ROOT_FOLDER = configs.ROOT_FOLDER
MODEL_ROOT = configs.MODEL_ROOT
RESULTS_FOLDER = configs.RESULTS_FOLDER
# Default params:
expt_name = "treatment_effects"
# EDIT ME! ######################################################################################
# Optimal network parameters to load for testing!
configs = [
configs.load_optimal_parameters('rnn_propensity_weighted',
expt_name,
add_net_name=True)
]
##################################################################################################
if __name__ == "__main__":
logging.basicConfig(format='%(levelname)s:%(message)s', level=logging.INFO)
# Setup tensorflow
tf_device = 'gpu'
if tf_device == "cpu":
tf_config = tf.ConfigProto(log_device_placement=False, device_count={'GPU': 0})
else: