Table of Contents
One of the best ways to get a feel for betway电竞deep RL is to run the algorithms and see how they perform on different tasks.The Spinning Up code library makes small-scale (local) experiments easy to do,and in this section,we'll discuss two ways to run them: either from the command line,or through function calls in scripts.
Spinning Up ships with
spinup/run.py,a convenient tool that lets you easily launch any algorithm (with any choices of hyperparameters) from the command line.It also serves as a thin wrapper over the utilities for watching trained policies and plotting,although we will not discuss that functionality on this page (for those details,see the pages onexperiment outputsandplotting).
The standard way to run a Spinning Up algorithm from the command line is
python -m spinup.run [algo name] [experiment flags]
python -m spinup.run ppo --env Walker2d-v2 --exp_name walker
You Should Know
If you are using ZShell: ZShell interprets square brackets as special characters.Spinning Up uses square brackets in a few ways for command line arguments;make sure to escape them,or try the solution recommendedhereif you want to escape them by default.
Detailed Quickstart Guide
python -m spinup.run ppo --exp_name ppo_ant --env Ant-v2 --clip_ratio 0.1 0.2 --hid[h] [32,32] [64,32] --act tf.nn.tanh --seed 0 10 20 --dt --data_dir path/to/data
runs PPO in the
Ant-v2Gym environment,with various settings controlled by the flags.
actare flags to set some algorithm hyperparameters.You can provide multiple values for hyperparameters to run multiple experiments.Check the docs to see what hyperparameters you can set (click here for thePPO documentation).
actarespecial shortcut flagsfor setting the hidden sizes and activation function for the neural networks trained by the algorithm.
seedflag sets the seed for the random number generator.RL algorithms have high variance,so try multiple seeds to get a feel for how performance varies.
dtflag ensures that the save directory names will have timestamps in them (otherwise they don't,unless you set
data_dirflag allows you to set the save folder for results.The default value is set by
spinup/user_config.py,which will be a subfolder
spinningupfolder (unless you change it).
Save directory namesare based on
exp_nameand any flags which have multiple values.Instead of the full flag,a shorthand will appear in the directory name.Shorthands can be provided by the user in square brackets after the flag,like
--hid[h];otherwise,shorthands are substrings of the flag (
cli).To illustrate,the save directory for the run with
Every hyperparameter in every algorithm can be controlled directly from the command line.If
kwargis a valid keyword arg for the function call of an algorithm,you can set values for it with the flag
--kwarg.To find out what keyword args are available,see either the docs page for an algorithm,or try
python -m spinup.run [algo name] --help
to see a readout of the docstring.
You Should Know
Values pass through
eval()before being used,so you can describe some functions and objects directly from the command line.For example:
python -m spinup.run ppo --env Walker2d-v2 --exp_name walker --act tf.nn.elu
tf.nn.eluas the activation function.
You Should Know
There's some nice handling for kwargs that take dict values.Instead of having to provide
--key dict(v1=value_1, v2=value_2)
you can give
--key:v1 value_1 --key:v2 value_2
to get the same result.
You can launch multiple experiments,to be executedin series,by simply providing more than one value for a given argument.(An experiment for each possible combination of values will be launched.)
For example,to launch otherwise-equivalent runs with different random seeds (0,10,and 20),do:
python -m spinup.run ppo --env Walker2d-v2 --exp_name walker --seed 0 10 20
Experiments don't launch in parallel because they soak up enough resources that executing several at the same time wouldn't get a speedup.
A few flags receive special treatment.
string.The name of an environment in the OpenAI Gym.All Spinning Up algorithms are implemented as functions that accept
env_fnas an argument,where
env_fnmust be a callable function that builds a copy of the RL environment.Since the most common use case is Gym environments,though,all of which are built through
gym.make(env_name),we allow you to just specify
envfor short) at the command line,which gets converted to a lambda-function that builds the correct gym environment.
Some algorithm arguments are relatively long,and we enabled shortcuts for them:
list of ints.Sets the sizes of the hidden layers in the neural networks (policies and value functions).
tf op.The activation function for the neural networks in the actor and critic.
These flags are valid for all current Spinning Up algorithms.
These flags are not hyperparameters of any algorithm,but change the experimental configuration in some way.
int.If this flag is set,the experiment is launched with this many processes,one per cpu,connected by MPI.Some algorithms are amenable to this sort of parallelization but not all.An error will be raised if you try setting
num_cpu> 1 for an incompatible algorithm.You can also set
--num_cpu auto,which will automatically use as many CPUs as are available on the machine.
string.The experiment name.This is used in naming the save directory for each experiment.The default is "cmd" + [algo name].
path.Set the base save directory for this experiment or set of experiments.If none is given,the
spinup/user_config.pywill be used.
bool.Include date and time in the name for the save directory of the experiment.
Results for a 必威电竞particular experiment (a single run of a configuration of hyperparameters) are stored in
data_diris the value of the
--data_dirflag (defaults to
--data_diris not given),
YY-MM-DD_timestamp if the
--datestampflag is raised,otherwise nothing,
YY-MM-DD_HH-MM-SS-timestamp if the
--datestampflag is raised,otherwise nothing,
suffixis a special string based on the experiment hyperparameters.
Suffixes are only included if you run multiple experiments at once,and they only include references to hyperparameters that differ across experiments,except for random seed.The goal is to make sure that results for similar experiments (ones which share all params except seed) are grouped in the same folder.
Suffixes are constructed by combiningshorthandsfor hyperparameters with their values,where a shorthand is either 1) constructed automatically from the hyperparameter name or 2) supplied by the user.The user can supply a shorthand by writing in square brackets after the kwarg flag.
python -m spinup.run ddpg --env Hopper-v2 --hid[h]  [128,128] --act tf.nn.tanh tf.nn.relu
--hidflag is given auser-supplied shorthand,
--actflag is not given a shorthand by the user,so one will be constructed for it automatically.
The suffixes produced in this case are:
_h128-128_ac-actrelu _h128-128_ac-acttanh _h300_ac-actrelu _h300_ac-acttanh
Note that the
hwas given by the user.the
ac-actshorthand was constructed from
ac_kwargs:activation(the true name for the
You Don't Actually Need to Know This One
Each individual algorithm is located in a file
spinup/algos/ALGO_NAME/ALGO_NAME.py,and these files can be run directly from the command line with a limited set of arguments (some of which differ from what's available to
spinup/run.py).The command line support in the individual algorithm files is essentially vestigial,however,and this isnota recommended way to perform experiments.
This documentation page will not describe those command line calls,and willonlydescribe calls through
Each algorithm is implemented as a python function,which can be imported directly from the
>>>from spinup import ppo
See the documentation page for each algorithm for a complete account of possible arguments.These methods can be used to set up specialized custom experiments,for example:
from spinup import ppo import tensorflow as tf import gym env_fn = lambda : gym.make('LunarLander-v2') ac_kwargs = dict(hidden_sizes=[64,64], activation=tf.nn.relu) logger_kwargs = dict(output_dir='path/to/output_dir', exp_name='experiment_name') ppo(env_fn=env_fn, ac_kwargs=ac_kwargs, steps_per_epoch=5000, epochs=250, logger_kwargs=logger_kwargs)
It's often useful in machine learning research to run the same algorithm with many possible hyperparameters.Spinning Up ships with a simple tool for facilitating this,calledExperimentGrid.
Consider the example in
1 2 3 4 5 6 7 8 910111213141516171819
from spinup.utils.run_utils import ExperimentGrid from spinup import ppo import tensorflow as tf if __name__ == '__main__': import argparse parser = argparse.ArgumentParser() parser.add_argument('--cpu', type=int, default=4) parser.add_argument('--num_runs', type=int, default=3) args = parser.parse_args() eg = ExperimentGrid(name='ppo-bench') eg.add('env_name', 'CartPole-v0', '', True) eg.add('seed', [10*i for i in range(args.num_runs)]) eg.add('epochs', 10) eg.add('steps_per_epoch', 4000) eg.add('ac_kwargs:hidden_sizes', [(32,), (64,64)], 'hid') eg.add('ac_kwargs:activation', [tf.tanh, tf.nn.relu], '') eg.run(ppo, num_cpu=args.cpu)
After making the ExperimentGrid object,parameters are added to it with
eg.add(param_name, values, shorthand, in_name)
in_nameforces a parameter to appear in the experiment name,even if it has the same value across all experiments.
After all parameters have been added,
runs all experiments in the grid (one experiment per valid configuration),by providing the configurations as kwargs to the function
ExperimentGrid.runuses a function namedcall_experimentto launch
**run_kwargsspecify behaviors for
call_experiment.Seethe documentation pagefor details.
Except for the absence of shortcut kwargs (you can't use
ExperimentGrid),the basic behavior of
ExperimentGridis the same as running things from the command line.(In fact,
ExperimentGridunder the hood.)