Related
I am trying to create a batched environment version of an SAC agent example from the Tensorflow Agents library, the original code can be found here. I am also using a custom environment.
I am pursuing a batched environment setup in order to better leverage GPU resources in order to speed up training. My understanding is that by passing batches of trajectories to the GPU, there will be less overhead incurred when passing data from the host (CPU) to the device (GPU).
My custom environment is called SacEnv, and I attempt to create a batched environment like so:
py_envs = [SacEnv() for _ in range(0, batch_size)]
batched_env = batched_py_environment.BatchedPyEnvironment(envs=py_envs)
tf_env = tf_py_environment.TFPyEnvironment(batched_env)
My hope is that this will create a batched environment consisting of a 'batch' of non-batched environments. However I am receiving the following error when running the code:
ValueError: Cannot assign value to variable ' Accumulator:0': Shape mismatch.The variable shape (1,), and the assigned value shape (32,) are incompatible.
with the stack trace:
Traceback (most recent call last):
File "/home/gary/Desktop/code/sac_test/sac_main2.py", line 370, in <module>
app.run(main)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/absl/app.py", line 312, in run
_run_main(main, args)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/absl/app.py", line 258, in _run_main
sys.exit(main(argv))
File "/home/gary/Desktop/code/sac_test/sac_main2.py", line 366, in main
train_eval(FLAGS.root_dir)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/gin/config.py", line 1605, in gin_wrapper
utils.augment_exception_message_and_reraise(e, err_str)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/gin/utils.py", line 41, in augment_exception_message_and_reraise
raise proxy.with_traceback(exception.__traceback__) from None
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/gin/config.py", line 1582, in gin_wrapper
return fn(*new_args, **new_kwargs)
File "/home/gary/Desktop/code/sac_test/sac_main2.py", line 274, in train_eval
results = metric_utils.eager_compute(
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/gin/config.py", line 1605, in gin_wrapper
utils.augment_exception_message_and_reraise(e, err_str)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/gin/utils.py", line 41, in augment_exception_message_and_reraise
raise proxy.with_traceback(exception.__traceback__) from None
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/gin/config.py", line 1582, in gin_wrapper
return fn(*new_args, **new_kwargs)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/eval/metric_utils.py", line 163, in eager_compute
common.function(driver.run)(time_step, policy_state)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tensorflow/python/util/traceback_utils.py", line 153, in error_handler
raise e.with_traceback(filtered_tb) from None
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/drivers/dynamic_episode_driver.py", line 211, in run
return self._run_fn(
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/utils/common.py", line 188, in with_check_resource_vars
return fn(*fn_args, **fn_kwargs)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/drivers/dynamic_episode_driver.py", line 238, in _run
tf.while_loop(
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/drivers/dynamic_episode_driver.py", line 154, in loop_body
observer_ops = [observer(traj) for observer in self._observers]
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/drivers/dynamic_episode_driver.py", line 154, in <listcomp>
observer_ops = [observer(traj) for observer in self._observers]
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/metrics/tf_metric.py", line 93, in __call__
return self._update_state(*args, **kwargs)
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/metrics/tf_metric.py", line 81, in _update_state
return self.call(*arg, **kwargs)
ValueError: in user code:
File "/home/gary/anaconda3/envs/py39/lib/python3.9/site-packages/tf_agents/metrics/tf_metrics.py", line 176, in call *
self._return_accumulator.assign(
ValueError: Cannot assign value to variable ' Accumulator:0': Shape mismatch.The variable shape (1,), and the assigned value shape (32,) are incompatible.
In call to configurable 'eager_compute' (<function eager_compute at 0x7fa4d6e5e040>)
In call to configurable 'train_eval' (<function train_eval at 0x7fa4c8622dc0>)
I have dug through the tf_metric.py code to try and understand the error, however I have been unsuccessful. A related issue was solved when I added the batch size (32) to the initializer for the AverageReturnMetric instance, and this issue seems related.
The full code is:
# coding=utf-8
# Copyright 2020 The TF-Agents Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# Lint as: python2, python3
r"""Train and Eval SAC.
All hyperparameters come from the SAC paper
https://arxiv.org/pdf/1812.05905.pdf
To run:
```bash
tensorboard --logdir $HOME/tmp/sac/gym/HalfCheetah-v2/ --port 2223 &
python tf_agents/agents/sac/examples/v2/train_eval.py \
--root_dir=$HOME/tmp/sac/gym/HalfCheetah-v2/ \
--alsologtostderr
\```
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from sac_env import SacEnv
import os
import time
from absl import app
from absl import flags
from absl import logging
import gin
from six.moves import range
import tensorflow as tf # pylint: disable=g-explicit-tensorflow-version-import
from tf_agents.agents.ddpg import critic_network
from tf_agents.agents.sac import sac_agent
from tf_agents.agents.sac import tanh_normal_projection_network
from tf_agents.drivers import dynamic_step_driver
#from tf_agents.environments import suite_mujoco
from tf_agents.environments import tf_py_environment
from tf_agents.environments import batched_py_environment
from tf_agents.eval import metric_utils
from tf_agents.metrics import tf_metrics
from tf_agents.networks import actor_distribution_network
from tf_agents.policies import greedy_policy
from tf_agents.policies import random_tf_policy
from tf_agents.replay_buffers import tf_uniform_replay_buffer
from tf_agents.utils import common
from tf_agents.train.utils import strategy_utils
flags.DEFINE_string('root_dir', os.getenv('TEST_UNDECLARED_OUTPUTS_DIR'),
'Root directory for writing logs/summaries/checkpoints.')
flags.DEFINE_multi_string('gin_file', None, 'Path to the trainer config files.')
flags.DEFINE_multi_string('gin_param', None, 'Gin binding to pass through.')
FLAGS = flags.FLAGS
gpus = tf.config.list_physical_devices('GPU')
if gpus:
try:
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
logical_gpus = tf.config.experimental.list_logical_devices('GPU')
print(len(gpus), "Physical GPUs,", len(logical_gpus), "Logical GPUs")
except RuntimeError as e:
print(e)
#gin.configurable
def train_eval(
root_dir,
env_name='SacEnv',
# The SAC paper reported:
# Hopper and Cartpole results up to 1000000 iters,
# Humanoid results up to 10000000 iters,
# Other mujoco tasks up to 3000000 iters.
num_iterations=3000000,
actor_fc_layers=(256, 256),
critic_obs_fc_layers=None,
critic_action_fc_layers=None,
critic_joint_fc_layers=(256, 256),
# Params for collect
# Follow https://github.com/haarnoja/sac/blob/master/examples/variants.py
# HalfCheetah and Ant take 10000 initial collection steps.
# Other mujoco tasks take 1000.
# Different choices roughly keep the initial episodes about the same.
#initial_collect_steps=10000,
initial_collect_steps=2000,
collect_steps_per_iteration=1,
replay_buffer_capacity=31250, # 1000000 / 32
# Params for target update
target_update_tau=0.005,
target_update_period=1,
# Params for train
train_steps_per_iteration=1,
#batch_size=256,
batch_size=32,
actor_learning_rate=3e-4,
critic_learning_rate=3e-4,
alpha_learning_rate=3e-4,
td_errors_loss_fn=tf.math.squared_difference,
gamma=0.99,
reward_scale_factor=0.1,
gradient_clipping=None,
use_tf_functions=True,
# Params for eval
num_eval_episodes=30,
eval_interval=10000,
# Params for summaries and logging
train_checkpoint_interval=50000,
policy_checkpoint_interval=50000,
rb_checkpoint_interval=50000,
log_interval=1000,
summary_interval=1000,
summaries_flush_secs=10,
debug_summaries=False,
summarize_grads_and_vars=False,
eval_metrics_callback=None):
"""A simple train and eval for SAC."""
root_dir = os.path.expanduser(root_dir)
train_dir = os.path.join(root_dir, 'train')
eval_dir = os.path.join(root_dir, 'eval')
train_summary_writer = tf.compat.v2.summary.create_file_writer(
train_dir, flush_millis=summaries_flush_secs * 1000)
train_summary_writer.set_as_default()
eval_summary_writer = tf.compat.v2.summary.create_file_writer(
eval_dir, flush_millis=summaries_flush_secs * 1000)
eval_metrics = [
tf_metrics.AverageReturnMetric(buffer_size=num_eval_episodes),
tf_metrics.AverageEpisodeLengthMetric(buffer_size=num_eval_episodes)
]
global_step = tf.compat.v1.train.get_or_create_global_step()
with tf.compat.v2.summary.record_if(
lambda: tf.math.equal(global_step % summary_interval, 0)):
py_envs = [SacEnv() for _ in range(0, batch_size)]
batched_env = batched_py_environment.BatchedPyEnvironment(envs=py_envs)
tf_env = tf_py_environment.TFPyEnvironment(batched_env)
eval_py_envs = [SacEnv() for _ in range(0, batch_size)]
eval_batched_env = batched_py_environment.BatchedPyEnvironment(envs=eval_py_envs)
eval_tf_env = tf_py_environment.TFPyEnvironment(eval_batched_env)
time_step_spec = tf_env.time_step_spec()
observation_spec = time_step_spec.observation
action_spec = tf_env.action_spec()
strategy = strategy_utils.get_strategy(tpu=False, use_gpu=True)
with strategy.scope():
actor_net = actor_distribution_network.ActorDistributionNetwork(
observation_spec,
action_spec,
fc_layer_params=actor_fc_layers,
continuous_projection_net=tanh_normal_projection_network
.TanhNormalProjectionNetwork)
critic_net = critic_network.CriticNetwork(
(observation_spec, action_spec),
observation_fc_layer_params=critic_obs_fc_layers,
action_fc_layer_params=critic_action_fc_layers,
joint_fc_layer_params=critic_joint_fc_layers,
kernel_initializer='glorot_uniform',
last_kernel_initializer='glorot_uniform')
tf_agent = sac_agent.SacAgent(
time_step_spec,
action_spec,
actor_network=actor_net,
critic_network=critic_net,
actor_optimizer=tf.compat.v1.train.AdamOptimizer(
learning_rate=actor_learning_rate),
critic_optimizer=tf.compat.v1.train.AdamOptimizer(
learning_rate=critic_learning_rate),
alpha_optimizer=tf.compat.v1.train.AdamOptimizer(
learning_rate=alpha_learning_rate),
target_update_tau=target_update_tau,
target_update_period=target_update_period,
td_errors_loss_fn=td_errors_loss_fn,
gamma=gamma,
reward_scale_factor=reward_scale_factor,
gradient_clipping=gradient_clipping,
debug_summaries=debug_summaries,
summarize_grads_and_vars=summarize_grads_and_vars,
train_step_counter=global_step)
tf_agent.initialize()
# Make the replay buffer.
replay_buffer = tf_uniform_replay_buffer.TFUniformReplayBuffer(
data_spec=tf_agent.collect_data_spec,
batch_size=batch_size,
max_length=replay_buffer_capacity,
device="/device:GPU:0")
replay_observer = [replay_buffer.add_batch]
train_metrics = [
tf_metrics.NumberOfEpisodes(),
tf_metrics.EnvironmentSteps(),
tf_metrics.AverageReturnMetric(
buffer_size=num_eval_episodes, batch_size=tf_env.batch_size),
tf_metrics.AverageEpisodeLengthMetric(
buffer_size=num_eval_episodes, batch_size=tf_env.batch_size),
]
eval_policy = greedy_policy.GreedyPolicy(tf_agent.policy)
initial_collect_policy = random_tf_policy.RandomTFPolicy(
tf_env.time_step_spec(), tf_env.action_spec())
collect_policy = tf_agent.collect_policy
train_checkpointer = common.Checkpointer(
ckpt_dir=train_dir,
agent=tf_agent,
global_step=global_step,
metrics=metric_utils.MetricsGroup(train_metrics, 'train_metrics'))
policy_checkpointer = common.Checkpointer(
ckpt_dir=os.path.join(train_dir, 'policy'),
policy=eval_policy,
global_step=global_step)
rb_checkpointer = common.Checkpointer(
ckpt_dir=os.path.join(train_dir, 'replay_buffer'),
max_to_keep=1,
replay_buffer=replay_buffer)
train_checkpointer.initialize_or_restore()
rb_checkpointer.initialize_or_restore()
initial_collect_driver = dynamic_step_driver.DynamicStepDriver(
tf_env,
initial_collect_policy,
observers=replay_observer + train_metrics,
num_steps=initial_collect_steps)
collect_driver = dynamic_step_driver.DynamicStepDriver(
tf_env,
collect_policy,
observers=replay_observer + train_metrics,
num_steps=collect_steps_per_iteration)
if use_tf_functions:
initial_collect_driver.run = common.function(initial_collect_driver.run)
collect_driver.run = common.function(collect_driver.run)
tf_agent.train = common.function(tf_agent.train)
if replay_buffer.num_frames() == 0:
# Collect initial replay data.
logging.info(
'Initializing replay buffer by collecting experience for %d steps '
'with a random policy.', initial_collect_steps)
initial_collect_driver.run()
results = metric_utils.eager_compute(
eval_metrics,
eval_tf_env,
eval_policy,
num_episodes=num_eval_episodes,
train_step=global_step,
summary_writer=eval_summary_writer,
summary_prefix='Metrics',
)
if eval_metrics_callback is not None:
eval_metrics_callback(results, global_step.numpy())
metric_utils.log_metrics(eval_metrics)
time_step = None
policy_state = collect_policy.get_initial_state(tf_env.batch_size)
timed_at_step = global_step.numpy()
time_acc = 0
# Prepare replay buffer as dataset with invalid transitions filtered.
def _filter_invalid_transition(trajectories, unused_arg1):
return ~trajectories.is_boundary()[0]
dataset = replay_buffer.as_dataset(
sample_batch_size=batch_size,
num_steps=2).unbatch().filter(
_filter_invalid_transition).batch(batch_size).prefetch(5)
# Dataset generates trajectories with shape [Bx2x...]
iterator = iter(dataset)
def train_step():
experience, _ = next(iterator)
return tf_agent.train(experience)
if use_tf_functions:
train_step = common.function(train_step)
global_step_val = global_step.numpy()
while global_step_val < num_iterations:
start_time = time.time()
time_step, policy_state = collect_driver.run(
time_step=time_step,
policy_state=policy_state,
)
for _ in range(train_steps_per_iteration):
train_loss = train_step()
time_acc += time.time() - start_time
global_step_val = global_step.numpy()
if global_step_val % log_interval == 0:
logging.info('step = %d, loss = %f', global_step_val,
train_loss.loss)
steps_per_sec = (global_step_val - timed_at_step) / time_acc
logging.info('%.3f steps/sec', steps_per_sec)
tf.compat.v2.summary.scalar(
name='global_steps_per_sec', data=steps_per_sec, step=global_step)
timed_at_step = global_step_val
time_acc = 0
for train_metric in train_metrics:
train_metric.tf_summaries(
train_step=global_step, step_metrics=train_metrics[:2])
if global_step_val % eval_interval == 0:
results = metric_utils.eager_compute(
eval_metrics,
eval_tf_env,
eval_policy,
num_episodes=num_eval_episodes,
train_step=global_step,
summary_writer=eval_summary_writer,
summary_prefix='Metrics',
)
if eval_metrics_callback is not None:
eval_metrics_callback(results, global_step_val)
metric_utils.log_metrics(eval_metrics)
if global_step_val % train_checkpoint_interval == 0:
train_checkpointer.save(global_step=global_step_val)
if global_step_val % policy_checkpoint_interval == 0:
policy_checkpointer.save(global_step=global_step_val)
if global_step_val % rb_checkpoint_interval == 0:
rb_checkpointer.save(global_step=global_step_val)
return train_loss
def main(_):
tf.compat.v1.enable_v2_behavior()
logging.set_verbosity(logging.INFO)
gin.parse_config_files_and_bindings(FLAGS.gin_file, FLAGS.gin_param)
train_eval(FLAGS.root_dir)
if __name__ == '__main__':
flags.mark_flag_as_required('root_dir')
app.run(main)
What is the appropriate way to create a batched environment for a custom, non-batched environment? I can share my custom environment, but I don't believe the issue lies there as the code works fine when using batch sizes of 1.
Also, any tips on increasing GPU utilization in reinforcement learning scenarios would be greatly appreciated. I have examined examples of using tensorboard-profiler to profile GPU utilization, but it seems these require callbacks and a fit function, which doesn't seem to be applicable in RL use-cases.
It turns out I neglected to pass batch_size when initializing the AverageReturnMetric and AverageEpisodeLengthMetric instances.
I have a working python script based on pandas.
Converting a similar script into a .exe worked at my computer at work. Unfortunately this isn't the case for my computer at home. I tried pyinstaller and py2exe and both bring up this error.
It seems to me that the conversion puts up a number of errors(I already fixed some of them), so it's not ultimately about the datetime issue I think.
import pandas as pd
import os
import glob
from datetime import datetime
import shutil
import os.path
try:
parentfolder = os.path.dirname(__file__)
parentfolder = os.path.abspath(os.path.join(parentfolder, '..'))#parentfolder der skriptdatei
except NameError: # We are the main py2exe script, not a module
import sys
parentfolder = os.path.dirname(sys.argv[0])
parentfolder = os.path.abspath(os.path.join(parentfolder, '..'))#parentfolder der skriptdatei
today = datetime.now()
day1 = today.strftime("%d-%m-%Y")
time1= today.strftime("%d-%m-%Y_%H-%M-%S")
day1=day1+'_cleaned'
logname="logfile_" + time1 + ".txt"
resultfolder=os.path.join(parentfolder, day1)
logfile = os.path.join(resultfolder, logname)
if os.path.exists(resultfolder):
shutil.rmtree(resultfolder) #deletes folder and all subfolders
os.makedirs(resultfolder)
pd.set_option('display.max_columns', 5)
pd.set_option('display.max_colwidth', 99)
f = open(logfile, "w")
f.close()
all_files = glob.glob(parentfolder + "/*.xls")
filecounter=0
first_run_counter=0
first_day_counter=0
for filename in all_files:
file_name=(os.path.splitext(os.path.basename(filename))[0])
writepath = os.path.join(resultfolder, '{}.xlsx'.format(str(file_name)+"_py"))
writer = pd.ExcelWriter(writepath, engine = 'xlsxwriter')
with open(logfile, "a") as file:
file.write("{} \n".format(str(file_name)))
filecounter += 1
if filecounter > 1:
print("WARNING, JUST CONVERT 1 FILE")
break
list1 = []
dfs_by_day= []
df = pd.read_excel(filename,header=None,parse_dates=False)#ohne header einlesen ,decimal=","
#df=df.convert_dtypes(convert_string=True)
df_help=df.copy()
df_help[1] = df_help[1].astype(str)
df_help[0] = df_help[0].astype(str)
#####datei ordnen,filtern etc
df.dropna(axis=0,how='any',thresh=None,subset=None,inplace=True)#löscht zeilen mit leeren zellen
df.drop_duplicates(inplace=True) #dropt auch doppelte header?!
df.reset_index(drop=True, inplace=True)
new_header = df.iloc[0] #grab the first row for the header
df = df[1:] #take the data less the header row
df.columns = new_header#nimmt 2, reihe als header
df = df.sort_values(['Date (MM/DD/YYYY)','Time (HH:mm:ss)'], ascending=[True,True])
df.reset_index(drop=True, inplace=True)
df.rename(columns={'Date (MM/DD/YYYY)':'Date (DD/MM/YYYY)'}, inplace=True)
#df['Date (DD/MM/YYYY)']=df['Date (DD/MM/YYYY)'].astype(str)#WICHTIG! datumsangabe unabhängig von / oder . machen
#df['Date (DD/MM/YYYY)'] = df['Date (DD/MM/YYYY)'].str.replace('/','.')#/ mit . ersetzen
df_help2=df.copy() #deepcopy vom noch nicht datetime, aber getrimmten dataframe
#################################################################### datei in tage aufspalten
##df_help2['Date (DD/MM/YYYY)'] = pd.to_datetime(df_help2['Date (DD/MM/YYYY)'],format='%d.%m.%Y')#EVTL FORMAT EINFÜGEN ,format='%d.%m.%Y'
df_help2['next day'] = (df_help2['Date (DD/MM/YYYY)'].diff()).dt.days > 0 #ob neue zeile=neuer tag
###############datumsangabe unabhängig von / oder . machen
for i in range(df_help2.shape[0]):
if df_help2.at[i,'next day'] == True:
list1.append(i)
#spaltalgorithmus gesamtfile in tage
l_mod = [0] + list1 + [df.shape[0]]
dfs_by_day = [df.iloc[l_mod[n]:l_mod[n+1]] for n in range(len(l_mod)-1)]
################################################################# tage in runs aufspalten
for j in dfs_by_day:
memo=0
run_counter=1
df1 = j
df1=df1.reset_index(drop=True)
df_help4 = df1.iloc[0:1,0:2].reset_index(drop=True).copy()
df1['Date (DD/MM/YYYY)'] = df1['Date (DD/MM/YYYY)'].dt.strftime('%d.%m.%Y')
list3=[]
dfdate= str(df1.at[0,'Date (DD/MM/YYYY)'])
print(dfdate)
df_help3=df1.copy() #deepcopy für tageszeitanalyse/runs
df_help3['Time (HH:mm:ss)'] = pd.to_datetime(df_help3['Time (HH:mm:ss)'],format='%H:%M:%S')
df_help3['next run'] = (df_help3['Time (HH:mm:ss)'].diff()).dt.seconds > 2000
df_help3.reset_index(drop=True, inplace=True)
for i in range(df_help3.shape[0]):
if df_help3.at[i,'next run'] == True:
list3.append(i)
###algorithmus spaltet tag in runs auf
l_mod2 = [0] + list3 + [df1.shape[0]]
dfs_by_run = [df1.iloc[l_mod2[n]:l_mod2[n+1]] for n in range(len(l_mod2)-1)]
for k in dfs_by_run:
df_run = k
df_run['Depth m'] = pd.to_numeric(df_run['Depth m'])
df_run['depth rounded'] = df_run['Depth m'].astype(int) #rundet
df_run=df_run.reset_index(drop=True)
df_run = df_run.drop_duplicates(subset=['depth rounded'], keep='last')#letzter wert
del df_run['depth rounded']
df_run=df_run.dropna(axis=0,how='any',thresh=2)
df_run=df_run.reset_index(drop=True)
run_name = str(dfdate) +'_run' + str(run_counter)
#####sensortoresultfile
if first_run_counter==0:
last_df=df_run.copy()
last_df=last_df[0:0]
last_df=last_df.append(df_run)
first_run_counter+=1
with open(logfile, "a") as file:
file.write("{0} has {1} last measurement(s) \n".format(run_name,df_run.shape[0]))
run_counter+=1
#alle daten raw aber mit sensor und header pro tag
df_help4['Time (HH:mm:ss)'] = df_help4['Time (HH:mm:ss)'].astype(str)
df_help4['Date (DD/MM/YYYY)'] = df_help4['Date (DD/MM/YYYY)'].astype(str)
for i in range(df_help.shape[0]):
if df_help4.at[0,'Date (DD/MM/YYYY)'] == df_help.at[i,0]:
if df_help4.at[0,'Time (HH:mm:ss)'] == df_help.at[i,1]:
memo=i
break
for n in reversed(list(range(memo))):
if df_help.at[n,3] == 'SENSOR SERIAL NUMBER:':
sensor_info=df_help.iloc[n:n+1,:]
sensor_info.reset_index(drop=True,inplace=True)
break
sensor_info.at[0,0:2]='-'
df1 = df1.columns.to_frame().T.append(df1, ignore_index=True)#fügt header als zeile ganz oben hinzu
df1.columns = range(len(df1.columns))#header neu 0 bis n
if first_day_counter==0:
raw_df=df1.copy()
raw_df=raw_df[0:0]
sensor_info.columns= range(len(df1.columns))
df1 = pd.concat([df1.iloc[:(0)], sensor_info, df1.iloc[0:]]).reset_index(drop=True)
raw_df=raw_df.append(df1)
first_day_counter += 1
last_df.to_excel(writer, sheet_name='{}'.format("last"),header=False, index = False)
#raw_df['Date (DD/MM/YYYY)'] = raw_df['Date (DD/MM/YYYY)'].dt.strftime('%d.%m.%Y')
raw_df.to_excel(writer, sheet_name='{}'.format("raw"),header=False, index = False)
writer.save()
with open(logfile, "a") as file:
file.write("total number of last measurements: {} \n".format(last_df.shape[0]))
file.write("total number of raw measurements: {} \n".format(raw_df.shape[0]))
f.close()
error:
Traceback (most recent call last):
File "tsk-py-convert.py", line 95, in <module>
File "pandas\core\generic.pyc", line 5458, in __getattr__
File "pandas\core\accessor.pyc", line 180, in __get__
File "pandas\core\indexes\accessors.pyc", line 494, in __new__
AttributeError: Can only use .dt accessor with datetimelike values
Within spyder the code was using an old pandas version (0.23.4). My code doesn't seem to work with a new version. I had the latest pandas version pip installed on windows and now manually installed the version of anaconda (0.23.4).
I can now run the code thorugh cmd, IDLE and the .exe that is created with pyinstaller works!
I used Model_Average_optimizer and some problems occured:
the code is used more likely as model_average_optimizer_test.py.
I find this error is occured in init ModelAverageOptimizer to create local_step. And this problem is more likely some problems with device_setter.
But I don't know why. Could someone help me?
Have I written custom code: Yes
OS Platform and Distribution: Linux n10-044-067 4.4.0-33.bm.1-amd64 #1 SMP Thu, 22 Jun 2017 11:19:55 +0800 x86_64 GNU/Linux
TensorFlow installed from: Anaconda2.5, pip install tensorflow_gpu
TensorFlow version: 1.8.0
Bazel version: N/A
CUDA/cuDNN version: CUDA-9.0, cuDNN-7.1.2(installed in anaconda2.5 also)
GPU model and memory: GeForce GTX 1080
Exact command to reproduce:
~/anaconda2/bin/python test.py --server_hosts=localhost:12222 --worker_hosts=localhost:12223,localhost:12224 --job_name=worker --task_id=0
~/anaconda2/bin/python test.py --server_hosts=localhost:12222 --worker_hosts=localhost:12223,localhost:12224 --job_name=worker --task_id=1
~/anaconda2/bin/python test.py --server_hosts=localhost:12222 --worker_hosts=localhost:12223,localhost:12224 --job_name=server --task_id=0
You may notice the errors below is occured when one worker init class ModelAverageOptimizer
Traceback (most recent call last):
File "/data00/home/wupeihao/ma_test/src/rnnlm.py", line 112, in tf.app.run()
File "/data00/home/wupeihao/anaconda2/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 126, in run
_sys.exit(main(argv))
File "/data00/home/wupeihao/ma_test/src/rnnlm.py", line 109, in main
test()
File "/data00/home/wupeihao/ma_test/src/rnnlm.py", line 84, in test
interval_steps=3)
File "/data00/home/wupeihao/anaconda2/lib/python2.7/site-packages/tensorflow/contrib/opt/python/training/model_average_optimizer.py", line 139, in init name="local_step")
File "/data00/home/wupeihao/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 1317, in get_variable
constraint=constraint)
File "/data00/home/wupeihao/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 1079, in get_variable
constraint=constraint)
File "/data00/home/wupeihao/anaconda2/lib/python2.7/site-packages/tensorflow/python/ops/variable_scope.py", line 417, in get_variable
return custom_getter(**custom_getter_kwargs)
File "/data00/home/wupeihao/anaconda2/lib/python2.7/site-packages/tensorflow/contrib/opt/python/training/model_average_optimizer.py", line 92, in call
return getter(name, trainable, collections, *args, **kwargs)
TypeError: _true_getter() got multiple values for keyword argument 'dtype'
The code is below:
import os
import time
import json
import copy
import numpy as np
import tensorflow as tf
from tensorflow.python.framework import ops
from tensorflow.contrib.opt.python.training import model_average_optimizer
flags = tf.flags
flags.DEFINE_string("server_hosts", "", "Comma-separated list of hostname:port pairs")
flags.DEFINE_string("worker_hosts", "", "Comma-separated list of hostname:port pairs")
flags.DEFINE_string("job_name", "", "Either 'server' of 'worker'")
flags.DEFINE_integer("task_id", 0, "Task Id for Each workers")
FLAGS = flags.FLAGS
tf.logging.set_verbosity(tf.logging.INFO)
def workers_ps_creator(args):
ps_hosts = args.server_hosts.split(",")
worker_hosts = args.worker_hosts.split(",")
num_workers = len(worker_hosts)
cluster = tf.train.ClusterSpec({"ps": ps_hosts,"worker": worker_hosts})
gpu_options = tf.GPUOptions(allocator_type='BFC', allow_growth=True)
if args.job_name == "server":
server_def = tf.train.ServerDef(cluster=cluster.as_cluster_def(),
job_name='ps',
task_index=args.task_id,
default_session_config=tf.ConfigProto(gpu_options=gpu_options, device_count={"GPU":0}),
protocol="grpc")
elif args.job_name == "worker":
server_def = tf.train.ServerDef(cluster=cluster.as_cluster_def(),
job_name="worker",
task_index=args.task_id,
default_session_config = tf.ConfigProto(gpu_options=gpu_options),
protocol="grpc")
server = tf.train.Server(server_def)
return server, cluster, num_workers, gpu_options
def Model(opt):
if FLAGS.task_id == 0:
var_0 = tf.get_variable(initializer = 0.0, name = 'v0')
var_1 = tf.get_variable(initializer = 1.0, name = 'v1')
grads_0 = constant_op.constant(-1.0)
grads_1 = constant_op.constant(-1.0)
else:
var_0 = tf.get_variable(initializer = 7.0, name = 'v0')
var_1 = tf.get_variable(initializer = 8.0, name = 'v1')
grads_0 = constant_op.constant(-2.0)
grads_1 = constant_op.constant(-2.0)
train_op = opt.apply_gradients([[grads_0, var_0], [grads_1, var_1]],
global_step = tf.train.get_or_create_global_step())
return train_op
def test():
server, cluster, num_workers, gpu_options = workers_ps_creator(FLAGS)
if FLAGS.job_name == "server":
server.join()
elif FLAGS.job_name == "worker":
is_chief = (FLAGS.task_id == 0)
#Between-graph replication
worker_device = "/job:worker/task:%d" % (FLAGS.task_id)
ma_custom = model_average_optimizer.ModelAverageCustomGetter(worker_device=worker_device)
from tensorflow.python.training import device_setter
with tf.device(
device_setter.replica_device_setter(
cluster=cluster,
worker_device=worker_device,
ps_device="/job:ps")), \
tf.variable_scope("", custom_getter=ma_custom):
#create model
lr = tf.Variable(1, trainable=False)
opt = tf.train.GradientDescentOptimizer(lr)
sync_opt = model_average_optimizer.ModelAverageOptimizer(
opt=opt,
num_worker=num_workers,
ma_custom_getter=ma_custom,
is_chief=is_chief,
interval_steps=3)
tf.logging.info('model start')
train_model = Model(sync_opt)
tf.logging.info('model end')
ma_hook = sync_opt.make_session_run_hook()
sess_config = tf.ConfigProto(gpu_options=gpu_options)
sess_config.log_device_placement = False
sess_config.allow_soft_placement = True
all_hooks = [ma_hook]
tf.logging.info('Start Sess')
with tf.train.MonitoredTrainingSession(master=server.target,
is_chief=is_chief,
hooks=all_hooks) as sess:
tf.logging.info("is chief: %s, len: %s", is_chief, num_workers)
for i in range(4):
sess.run(train_op)
pp1 = sess.run(tf.get_default_graph().get_tensor_by_name('v0:0'))
pp2 = sess.run(tf.get_default_graph().get_tensor_by_name('v1:0'))
tf.logging.info("%d %.2f %.2f" % (FLAGS.task_id, pp1, pp2))
sv.stop()
tf.logging.info("done")
I know how to load data into Scrapy spider from external source when working localy. But I strugle to find any info on how to deploy this file to scrapinghub and what path to use there. Now i use this approach from SH documentation - enter link description here but recieve NONE object.
import pkgutil
class CodeSpider(scrapy.Spider):
name = "code"
allowed_domains = ["google.com.au"]
def start_requests(self, ):
f = pkgutil.get_data("project", "res/final.json")
a = json.loads(f.read())
Thanks.
My setup file
from setuptools import setup, find_packages
setup(
name = 'project',
version = '1.0',
packages = find_packages(),
package_data = {'project': ['res/*.json']
},
entry_points = {'scrapy': ['settings = au_go.settings']},
zip_safe=False,
)
The error i got.
Traceback (most recent call last):
File "/usr/local/lib/python2.7/site-packages/scrapy/core/engine.py", line 127, in _next_request
request = next(slot.start_requests)
File "/tmp/unpacked-eggs/__main__.egg/au_go/spiders/code.py", line 16, in start_requests
a = json.loads(f.read())
AttributeError: 'NoneType' object has no attribute 'read'
From the traceback you supplied, I assume that your project files look like this:
au_go/
__init__.py
settings.py
res/
final.json
spiders/
__init__.py
code.py
scrapy.cfg
setup.py
With this assumption, the setup.py's package_data needs to refer to the package named au_go:
from setuptools import setup, find_packages
setup(
name = 'au_go',
version = '1.0',
packages = find_packages(),
package_data = {
'au_go': ['res/*.json']
},
entry_points = {'scrapy': ['settings = au_go.settings']},
zip_safe=False,
)
And then you can use pkgutil.get_data("au_go", "res/final.json").
Recently I want to try some open source solvers instead of CPLEX. I found that PICOS + zibopt may be a good choice. However, I can merely find instruction on how to make zibopt work with python under windows properly. I downloaded the windows libraries (.dll file) of scip, and I try to install python-zibopt according to the command "python setup.py install". The error " blockmemshell/memory.h no such file" always popped out. I felt that it is because my compiler, which is VS120COMNTOOL, doecn't find the scip solver. Is there any chance that I can make scip work under windows now?
Did you have a look at the current python interface of SCIP 3.1.0? It uses the library from the SCIP Optimization Suite so you don't have to link another LP solver to SCIP.
On Windows, please try this modified setup.py file:
import sys, os, readline, glob, platform
from distutils.core import setup
from distutils.extension import Extension
from Cython.Distutils import build_ext
from Cython.Build import cythonize
BASEDIR = os.path.dirname(os.path.abspath(__file__))
BASEDIR = os.path.dirname(BASEDIR)
BASEDIR = os.path.dirname(BASEDIR)
INCLUDEDIR = os.path.join(BASEDIR,'src')
BASEDIR = os.path.dirname(BASEDIR)
#identify compiler version
prefix = "MSC v."
i = sys.version.find(prefix)
if i == -1:
raise Exception('cannot determine compiler version')
i = i + len(prefix)
s, rest = sys.version[i:].split(" ", 1)
majorVersion = int(s[:-2]) - 6
minorVersion = int(s[2:3]) / 10.0
if platform.architecture()[0].find('64')>=0:
LIBDIR = os.path.join(BASEDIR,'vc'+str(majorVersion),'scip_spx','x64','Release')
else:
LIBDIR = os.path.join(BASEDIR,'vc'+str(majorVersion),'scip_spx','Release')
print('BASEDIR='+ BASEDIR)
print('INCLUDEDIR='+ INCLUDEDIR)
print('LIBDIR='+ LIBDIR)
def complete(text, state):
return (glob.glob(text+'*')+[None])[state]
readline.set_completer_delims(' \t\n;')
readline.parse_and_bind("tab: complete")
readline.set_completer(complete)
libscipopt = 'lib/libscipopt.so'
includescip = 'include/scip'
ext_modules = []
ext_modules += [Extension('pyscipopt.scip', [os.path.join('pyscipopt', 'scip.pyx')],
#extra_compile_args=['-g', '-O0', '-UNDEBUG'],
include_dirs=[INCLUDEDIR],
library_dirs=[LIBDIR],
#runtime_library_dirs=[os.path.abspath('lib')],
libraries=['spx', 'scip_spx'])]
#libraries=['scipopt', 'readline', 'z', 'gmp', 'ncurses', 'm'])]
setup(
name = 'pyscipopt',
version = '0.1',
description = 'wrapper for SCIP in Python',
author = 'Zuse Institute Berlin',
author_email = 'scip#zib.de',
license = 'MIT',
cmdclass = {'build_ext' : build_ext},
ext_modules = ext_modules,
packages=['pyscipopt']
)