mirror of
https://github.com/facebookresearch/ReAgent.git
synced 2026-05-17 12:40:39 +00:00
Upgrade ReAgent to use Python 3.8 (#415)
Summary: Pull Request resolved: https://github.com/facebookresearch/ReAgent/pull/415 Currently, we have some test failures (https://app.circleci.com/pipelines/github/facebookresearch/ReAgent/1460/workflows/ecc21254-779b-4a89-a40d-ea317e839d96/jobs/8655) because we miss some latest features. Reviewed By: MisterTea Differential Revision: D26977836 fbshipit-source-id: 9243d194ddf5c62895c9f1369830309c379fd7dd
This commit is contained in:
Zhengxing Chen
committed by
Ruiyang Xu
Ruiyang Xu
parent
ae12656790
commit
cae245749a
+149
-29
@@ -83,15 +83,17 @@ commands:
|
||||
name: Installing SDKs
|
||||
command: |
|
||||
mv ~/.bashrc ~/.bashrc.bk
|
||||
sudo apt-get update
|
||||
sudo apt-get install bc
|
||||
sudo apt-get install unzip
|
||||
sudo apt-get install zip
|
||||
curl -s "https://get.sdkman.io" | bash
|
||||
source "$HOME/.sdkman/bin/sdkman-init.sh"
|
||||
sdk version
|
||||
sdk install java 8.0.272.hs-adpt
|
||||
sdk install scala
|
||||
sdk install maven
|
||||
sdk install spark 2.4.6
|
||||
sudo apt-get update
|
||||
sudo apt-get install bc
|
||||
sdk install spark 3.1.1
|
||||
- run:
|
||||
name: Build preprocessing package
|
||||
command: |
|
||||
@@ -140,12 +142,16 @@ commands:
|
||||
steps:
|
||||
- run:
|
||||
command: |
|
||||
pyenv global 3.7.0
|
||||
pyenv install -v 3.8.1
|
||||
pyenv global 3.8.1
|
||||
- run:
|
||||
command: |
|
||||
pip install --upgrade pip
|
||||
pip install tox==3.20.1
|
||||
pip install --upgrade wheel setuptools
|
||||
sudo apt update
|
||||
sudo apt install cmake
|
||||
sudo apt install swig
|
||||
pip install --upgrade pip --progress-bar off
|
||||
pip install --upgrade wheel setuptools --progress-bar off
|
||||
pip install tox==3.20.1 --progress-bar off
|
||||
- when:
|
||||
condition: << parameters.install_gym >>
|
||||
steps:
|
||||
@@ -154,13 +160,13 @@ commands:
|
||||
steps:
|
||||
- run:
|
||||
command: |
|
||||
pip install -e .[gym,test] --pre -f https://download.pytorch.org/whl/nightly/cu101/torch_nightly.html
|
||||
pip install -e .[gym,test] --pre -f https://download.pytorch.org/whl/nightly/cu102/torch_nightly.html --progress-bar off
|
||||
- unless:
|
||||
condition: << parameters.is_ubuntu_gpu >>
|
||||
steps:
|
||||
- run:
|
||||
command: |
|
||||
sudo pip install -e .[gym,test] --pre -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
|
||||
sudo pip install -e .[gym,test] --pre -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html --progress-bar off
|
||||
|
||||
run_unittest:
|
||||
description: Run unittests, coverage and save results
|
||||
@@ -171,14 +177,14 @@ commands:
|
||||
- run:
|
||||
no_output_timeout: 30m
|
||||
command: |
|
||||
tox -vv -e << parameters.tox_env >>
|
||||
tox -v -e << parameters.tox_env >>
|
||||
bash <(curl -s https://codecov.io/bash)
|
||||
- run: python setup.py bdist_wheel
|
||||
- run: python setup.py -q bdist_wheel
|
||||
- store_artifacts:
|
||||
path: dist/reagent-0.1-py3-none-any.whl
|
||||
destination: reagent-0.1-py3-none-any.whl
|
||||
- store_test_results:
|
||||
path: .tox/py37/log/
|
||||
path: .tox/py38/log/
|
||||
|
||||
run_interrogate:
|
||||
description: Install and run interrogate
|
||||
@@ -186,16 +192,16 @@ commands:
|
||||
- run:
|
||||
name: Install interrogate
|
||||
command: |
|
||||
pip install interrogate
|
||||
pip install interrogate --progress-bar off
|
||||
- run:
|
||||
name: Run interrogate on reagent code base
|
||||
command: |
|
||||
interrogate -piImvv -f 15 reagent/
|
||||
|
||||
jobs:
|
||||
gpu_unittest:
|
||||
misc_unittest:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.1:201909-23
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CUDA_LAUNCH_BLOCKING: 1
|
||||
@@ -205,11 +211,47 @@ jobs:
|
||||
install_gym: false
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_unittest
|
||||
tox_env: circleci_misc_unittest
|
||||
|
||||
gym_unittest:
|
||||
gym_cpu_unittest:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.1:201909-23
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: large
|
||||
steps:
|
||||
- checkout_merge
|
||||
- pip_install:
|
||||
install_gym: false
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_gym_cpu_unittest
|
||||
|
||||
gym_replay_buffer_cpu_unittest_1:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: large
|
||||
steps:
|
||||
- checkout_merge
|
||||
- pip_install:
|
||||
install_gym: false
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_gym_replay_buffer_1_cpu_unittest
|
||||
|
||||
gym_replay_buffer_cpu_unittest_2:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: large
|
||||
steps:
|
||||
- checkout_merge
|
||||
- pip_install:
|
||||
install_gym: false
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_gym_replay_buffer_2_cpu_unittest
|
||||
|
||||
gym_gpu_unittest:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CUDA_LAUNCH_BLOCKING: 1
|
||||
@@ -219,11 +261,39 @@ jobs:
|
||||
install_gym: false
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_gym_unittest
|
||||
tox_env: circleci_gym_gpu_unittest
|
||||
|
||||
gym_replay_buffer_gpu_unittest_1:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CUDA_LAUNCH_BLOCKING: 1
|
||||
steps:
|
||||
- checkout_merge
|
||||
- pip_install:
|
||||
install_gym: false
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_gym_replay_buffer_1_gpu_unittest
|
||||
|
||||
gym_replay_buffer_gpu_unittest_2:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CUDA_LAUNCH_BLOCKING: 1
|
||||
steps:
|
||||
- checkout_merge
|
||||
- pip_install:
|
||||
install_gym: false
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_gym_replay_buffer_2_gpu_unittest
|
||||
|
||||
dqn_cartpole_e2e:
|
||||
docker:
|
||||
- image: circleci/python:3.7
|
||||
- image: circleci/python:3.8
|
||||
resource_class: large
|
||||
environment:
|
||||
- BASH_ENV: ~/.bashrc
|
||||
@@ -235,9 +305,9 @@ jobs:
|
||||
is_ubuntu_gpu: false
|
||||
- end_to_end_test
|
||||
|
||||
seq2slate_e2e:
|
||||
ranking_unittest:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.1:201909-23
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CUDA_LAUNCH_BLOCKING: 1
|
||||
@@ -247,11 +317,53 @@ jobs:
|
||||
install_gym: true
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_seq2slate_unittest
|
||||
tox_env: circleci_ranking_unittest
|
||||
|
||||
training_unittest:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CUDA_LAUNCH_BLOCKING: 1
|
||||
steps:
|
||||
- checkout_merge
|
||||
- pip_install:
|
||||
install_gym: true
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_training_unittest
|
||||
|
||||
prediction_unittest:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CUDA_LAUNCH_BLOCKING: 1
|
||||
steps:
|
||||
- checkout_merge
|
||||
- pip_install:
|
||||
install_gym: true
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_prediction_unittest
|
||||
|
||||
world_model_unittest:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CUDA_LAUNCH_BLOCKING: 1
|
||||
steps:
|
||||
- checkout_merge
|
||||
- pip_install:
|
||||
install_gym: true
|
||||
is_ubuntu_gpu: true
|
||||
- run_unittest:
|
||||
tox_env: circleci_world_model_unittest
|
||||
|
||||
sac_pendulum_e2e:
|
||||
docker:
|
||||
- image: circleci/python:3.7
|
||||
- image: circleci/python:3.8
|
||||
resource_class: large
|
||||
environment:
|
||||
- BASH_ENV: ~/.bashrc
|
||||
@@ -265,7 +377,7 @@ jobs:
|
||||
|
||||
sac_pendulum_e2e_gpu:
|
||||
machine:
|
||||
image: ubuntu-1604-cuda-10.1:201909-23
|
||||
image: ubuntu-1604-cuda-10.2:202012-01
|
||||
resource_class: gpu.medium
|
||||
environment:
|
||||
- CONFIG: reagent/workflow/sample_configs/sac_pendulum_offline.yaml
|
||||
@@ -301,7 +413,7 @@ jobs:
|
||||
|
||||
docstring_coverage:
|
||||
docker:
|
||||
- image: circleci/python:3.7
|
||||
- image: circleci/python:3.8
|
||||
resource_class: small
|
||||
steps:
|
||||
- checkout_merge
|
||||
@@ -310,12 +422,20 @@ jobs:
|
||||
workflows:
|
||||
build:
|
||||
jobs:
|
||||
- seq2slate_e2e
|
||||
- ranking_unittest
|
||||
- training_unittest
|
||||
- prediction_unittest
|
||||
- world_model_unittest
|
||||
- dqn_cartpole_e2e
|
||||
- sac_pendulum_e2e
|
||||
- sac_pendulum_e2e_gpu
|
||||
- gpu_unittest
|
||||
- gym_unittest
|
||||
- misc_unittest
|
||||
- gym_cpu_unittest
|
||||
- gym_gpu_unittest
|
||||
- gym_replay_buffer_cpu_unittest_1
|
||||
- gym_replay_buffer_cpu_unittest_2
|
||||
- gym_replay_buffer_gpu_unittest_1
|
||||
- gym_replay_buffer_gpu_unittest_2
|
||||
- rasp_test_linux
|
||||
- rasp_test_mac
|
||||
- docstring_coverage
|
||||
|
||||
@@ -7,7 +7,7 @@ ReAgent CLI & Python API
|
||||
^^^^^^^^^^^^^^^^^^^^^^^^
|
||||
|
||||
We have CLI to launch training & Python API to use programmatically, e.g., in your own script or Jupyter Notebook.
|
||||
To install this component, you will need to have Python 3.7+ installed on your system.
|
||||
To install this component, you will need to have Python 3.8+ installed on your system.
|
||||
If you don't have that, you can either install it via `pyenv <https://github.com/pyenv/pyenv>`_ or
|
||||
`conda <https://docs.conda.io/projects/conda/en/latest/index.html>`_. To verify that you have the right version,
|
||||
type the following command on your shell:
|
||||
@@ -24,7 +24,7 @@ Once you make sure you have the right version, you can simply clone this repo an
|
||||
cd ReAgent
|
||||
pip install ".[gym]"
|
||||
|
||||
# install nightly torch (change cpu to cu101/102 if fit)
|
||||
# install nightly torch (change cpu to cu102 if fit)
|
||||
pip install --pre torch torchvision -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
|
||||
|
||||
If you don't want need gym dependencies, you can remove :code:`[gym]`
|
||||
@@ -49,7 +49,7 @@ To build from source, you'll need JDK, Scala, & Maven. We will use `SDKMAN! <htt
|
||||
curl -s "https://get.sdkman.io" | bash
|
||||
source "$HOME/.sdkman/bin/sdkman-init.sh"
|
||||
sdk version
|
||||
sdk install java 8.0.265.hs-adpt
|
||||
sdk install java 8.0.272.hs-adpt
|
||||
sdk install scala
|
||||
sdk install maven
|
||||
|
||||
@@ -57,7 +57,7 @@ If you are testing locally, you can also install Spark
|
||||
|
||||
.. code-block:: bash
|
||||
|
||||
sdk install spark 2.4.6
|
||||
sdk install spark 3.1.1
|
||||
|
||||
Now, you can build our preprocessing JAR
|
||||
|
||||
|
||||
@@ -37,13 +37,13 @@
|
||||
provided
|
||||
</parquet.deps.scope>
|
||||
<scala.version>
|
||||
2.11.7
|
||||
2.12.10
|
||||
</scala.version>
|
||||
<scala.binary.version>
|
||||
2.11
|
||||
2.12
|
||||
</scala.binary.version>
|
||||
<spark.version>
|
||||
2.3.2
|
||||
3.1.1
|
||||
</spark.version>
|
||||
</properties>
|
||||
<dependencies>
|
||||
@@ -55,7 +55,7 @@
|
||||
scalatest_${scala.binary.version}
|
||||
</artifactId>
|
||||
<version>
|
||||
2.2.6
|
||||
3.2.5
|
||||
</version>
|
||||
<scope>
|
||||
test
|
||||
@@ -69,7 +69,7 @@
|
||||
jacoco-maven-plugin
|
||||
</artifactId>
|
||||
<version>
|
||||
0.8.5
|
||||
0.8.6
|
||||
</version>
|
||||
<scope>
|
||||
test
|
||||
@@ -97,7 +97,7 @@
|
||||
scalacheck_${scala.binary.version}
|
||||
</artifactId>
|
||||
<version>
|
||||
1.13.5
|
||||
1.14.1
|
||||
</version>
|
||||
<scope>
|
||||
test
|
||||
@@ -293,7 +293,7 @@
|
||||
scala-maven-plugin
|
||||
</artifactId>
|
||||
<version>
|
||||
3.2.2
|
||||
4.4.1
|
||||
</version>
|
||||
<executions>
|
||||
<execution>
|
||||
|
||||
@@ -128,10 +128,18 @@ object Timeline {
|
||||
sqlContext: SQLContext,
|
||||
config: TimelineConfiguration
|
||||
): Unit = {
|
||||
var filterTerminal = "HAVING next_state_features IS NOT NULL";
|
||||
var filterTerminal = "WHERE next_state_features IS NOT NULL";
|
||||
if (config.addTerminalStateRow) {
|
||||
filterTerminal = "";
|
||||
}
|
||||
var filterTimeLimit = "";
|
||||
if (config.timeWindowLimit != None) {
|
||||
if (filterTerminal == "") {
|
||||
filterTimeLimit = s"WHERE time_since_first <= ${config.timeWindowLimit.get}";
|
||||
} else {
|
||||
filterTimeLimit = s" AND time_since_first <= ${config.timeWindowLimit.get}";
|
||||
}
|
||||
}
|
||||
|
||||
val actionDataType =
|
||||
Helper.getDataTypes(sqlContext, config.inputTableName, List("action"))("action")
|
||||
@@ -193,23 +201,6 @@ object Timeline {
|
||||
case (acc, (k, v)) => s"${acc}, a.${k}"
|
||||
}
|
||||
|
||||
val timeLimitedSourceTable = config.timeWindowLimit
|
||||
.map { timeLimit =>
|
||||
s"""
|
||||
, time_limited_source_table AS (
|
||||
SELECT
|
||||
*,
|
||||
sequence_number - FIRST(sequence_number) OVER (
|
||||
PARTITION BY mdp_id
|
||||
ORDER BY mdp_id, sequence_number
|
||||
) AS time_since_first
|
||||
FROM source_table
|
||||
HAVING time_since_first <= ${timeLimit}
|
||||
)
|
||||
""".stripMargin
|
||||
}
|
||||
.getOrElse("")
|
||||
|
||||
val sourceTable = s"""
|
||||
WITH ${mdpFilter}
|
||||
source_table AS (
|
||||
@@ -225,15 +216,8 @@ object Timeline {
|
||||
${joinClause}
|
||||
a.ds BETWEEN '${config.startDs}' AND '${config.endDs}'
|
||||
)
|
||||
${timeLimitedSourceTable}
|
||||
""".stripMargin
|
||||
|
||||
val sourceTableName = config.timeWindowLimit
|
||||
.map { _ =>
|
||||
"time_limited_source_table"
|
||||
}
|
||||
.getOrElse("source_table")
|
||||
|
||||
val rewardColumnsQuery = rewardColumnDataTypes.foldLeft("") {
|
||||
case (acc, (k, v)) => s"${acc}, ${k}"
|
||||
}
|
||||
@@ -253,53 +237,59 @@ object Timeline {
|
||||
}
|
||||
|
||||
val sqlCommand = s"""
|
||||
${sourceTable}
|
||||
${sourceTable},
|
||||
joined_table AS (
|
||||
SELECT
|
||||
mdp_id,
|
||||
state_features,
|
||||
action,
|
||||
LEAD(action) OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
) AS next_action,
|
||||
action_probability
|
||||
${rewardColumnsQuery},
|
||||
LEAD(state_features) OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
) AS next_state_features,
|
||||
sequence_number,
|
||||
ROW_NUMBER() OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
) AS sequence_number_ordinal,
|
||||
COALESCE(LEAD(sequence_number) OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
), sequence_number) - sequence_number AS time_diff,
|
||||
sequence_number - FIRST(sequence_number) OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
) AS time_since_first
|
||||
${timelineJoinColumnsQuery}
|
||||
FROM source_table
|
||||
CLUSTER BY HASH(mdp_id, sequence_number)
|
||||
)
|
||||
SELECT
|
||||
mdp_id,
|
||||
state_features,
|
||||
action,
|
||||
LEAD(action) OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
) AS next_action,
|
||||
action_probability
|
||||
${rewardColumnsQuery},
|
||||
LEAD(state_features) OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
) AS next_state_features,
|
||||
sequence_number,
|
||||
ROW_NUMBER() OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
) AS sequence_number_ordinal,
|
||||
COALESCE(LEAD(sequence_number) OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
), sequence_number) - sequence_number AS time_diff,
|
||||
sequence_number - FIRST(sequence_number) OVER (
|
||||
PARTITION BY
|
||||
mdp_id
|
||||
ORDER BY
|
||||
mdp_id,
|
||||
sequence_number
|
||||
) AS time_since_first
|
||||
${timelineJoinColumnsQuery}
|
||||
FROM ${sourceTableName}
|
||||
*
|
||||
FROM joined_table
|
||||
${filterTerminal}
|
||||
CLUSTER BY HASH(mdp_id, sequence_number)
|
||||
${filterTimeLimit}
|
||||
""".stripMargin
|
||||
log.info("Executing query: ")
|
||||
log.info(sqlCommand)
|
||||
|
||||
+3
-2
@@ -12,12 +12,13 @@ import org.apache.spark.sql.functions.col
|
||||
import org.apache.spark.sql._
|
||||
import org.apache.spark.{SparkConf, SparkContext}
|
||||
import org.apache.spark.sql.types._
|
||||
import org.scalatest.{BeforeAndAfterAll, FunSuiteLike, Suite}
|
||||
import org.scalatest.{BeforeAndAfterAll, Suite}
|
||||
import org.scalatest.funsuite.AnyFunSuiteLike
|
||||
|
||||
import scala.collection.mutable
|
||||
import scala.math.abs
|
||||
|
||||
trait PipelineTester extends FunSuiteLike with BeforeAndAfterAll with TestLogging { this: Suite =>
|
||||
trait PipelineTester extends AnyFunSuiteLike with BeforeAndAfterAll with TestLogging { this: Suite =>
|
||||
|
||||
@transient private var _sparkContext: SparkContext = _
|
||||
def sparkContext: SparkContext = _sparkContext
|
||||
|
||||
@@ -10,8 +10,8 @@ import org.scalatest._
|
||||
import scala.collection.JavaConversions._
|
||||
import scala.util.Try
|
||||
|
||||
trait TestLogging extends BeforeAndAfterAll with BeforeAndAfterEach with TestLogger {
|
||||
this: Suite =>
|
||||
trait TestLogging extends BeforeAndAfterAll with BeforeAndAfterEach with TestLogger with TestSuiteMixin {
|
||||
this: TestSuite =>
|
||||
|
||||
private val logLayout = new EnhancedPatternLayout("%d{yy/MM/dd HH:mm:ss} %p %c{1}: %m%n")
|
||||
|
||||
@@ -72,6 +72,7 @@ trait TestLogging extends BeforeAndAfterAll with BeforeAndAfterEach with TestLog
|
||||
val scopes = test.scopes
|
||||
val text = test.text
|
||||
val tags = test.tags
|
||||
val pos = test.pos
|
||||
}
|
||||
|
||||
super.withFixture(wrappedTest)
|
||||
|
||||
@@ -27,9 +27,9 @@ model:
|
||||
- leaky_relu
|
||||
eval_parameters:
|
||||
calc_cpe_in_training: false
|
||||
replay_memory_size: 20000
|
||||
replay_memory_size: 100000
|
||||
train_every_ts: 1
|
||||
train_after_ts: 5000
|
||||
train_after_ts: 20000
|
||||
num_train_episodes: 30
|
||||
num_eval_episodes: 20
|
||||
passing_score_bar: 100.0
|
||||
|
||||
@@ -4,6 +4,7 @@ import logging
|
||||
import os
|
||||
import pprint
|
||||
import unittest
|
||||
import uuid
|
||||
from typing import Optional, Dict, Any
|
||||
|
||||
import numpy as np
|
||||
@@ -48,7 +49,7 @@ Format path to be: "configs/<env_name>/<model_name>_<env_name>_online.yaml."
|
||||
NOTE: These tests should ideally finish quickly (within 10 minutes) since they are
|
||||
unit tests which are run many times.
|
||||
"""
|
||||
REPLAY_BUFFER_GYM_TESTS = [
|
||||
REPLAY_BUFFER_GYM_TESTS_1 = [
|
||||
("Discrete CRR Cartpole", "configs/cartpole/discrete_crr_cartpole_online.yaml"),
|
||||
("Discrete DQN Cartpole", "configs/cartpole/discrete_dqn_cartpole_online.yaml"),
|
||||
("Discrete C51 Cartpole", "configs/cartpole/discrete_c51_cartpole_online.yaml"),
|
||||
@@ -58,6 +59,8 @@ REPLAY_BUFFER_GYM_TESTS = [
|
||||
"configs/open_gridworld/discrete_dqn_open_gridworld.yaml",
|
||||
),
|
||||
("SAC Pendulum", "configs/pendulum/sac_pendulum_online.yaml"),
|
||||
]
|
||||
REPLAY_BUFFER_GYM_TESTS_2 = [
|
||||
("Continuous CRR Pendulum", "configs/pendulum/continuous_crr_pendulum_online.yaml"),
|
||||
("TD3 Pendulum", "configs/pendulum/td3_pendulum_online.yaml"),
|
||||
("Parametric DQN Cartpole", "configs/cartpole/parametric_dqn_cartpole_online.yaml"),
|
||||
@@ -91,8 +94,16 @@ curr_dir = os.path.dirname(__file__)
|
||||
|
||||
class TestGym(HorizonTestBase):
|
||||
# pyre-fixme[16]: Module `parameterized` has no attribute `expand`.
|
||||
@parameterized.expand(REPLAY_BUFFER_GYM_TESTS)
|
||||
def test_replay_buffer_gym_cpu(self, name: str, config_path: str):
|
||||
@parameterized.expand(REPLAY_BUFFER_GYM_TESTS_1)
|
||||
def test_replay_buffer_gym_cpu_1(self, name: str, config_path: str):
|
||||
self._test_replay_buffer_gym_cpu(name, config_path)
|
||||
|
||||
# pyre-fixme[16]: Module `parameterized` has no attribute `expand`.
|
||||
@parameterized.expand(REPLAY_BUFFER_GYM_TESTS_2)
|
||||
def test_replay_buffer_gym_cpu_2(self, name: str, config_path: str):
|
||||
self._test_replay_buffer_gym_cpu(name, config_path)
|
||||
|
||||
def _test_replay_buffer_gym_cpu(self, name: str, config_path: str):
|
||||
logger.info(f"Starting {name} on CPU")
|
||||
self.run_from_config(
|
||||
run_test=run_test_replay_buffer,
|
||||
@@ -102,10 +113,20 @@ class TestGym(HorizonTestBase):
|
||||
logger.info(f"{name} passes!")
|
||||
|
||||
# pyre-fixme[16]: Module `parameterized` has no attribute `expand`.
|
||||
@parameterized.expand(REPLAY_BUFFER_GYM_TESTS)
|
||||
@parameterized.expand(REPLAY_BUFFER_GYM_TESTS_1)
|
||||
@pytest.mark.serial
|
||||
@unittest.skipIf(not torch.cuda.is_available(), "CUDA not available")
|
||||
def test_replay_buffer_gym_gpu(self, name: str, config_path: str):
|
||||
def test_replay_buffer_gym_gpu_1(self, name: str, config_path: str):
|
||||
self._test_replay_buffer_gym_gpu(name, config_path)
|
||||
|
||||
# pyre-fixme[16]: Module `parameterized` has no attribute `expand`.
|
||||
@parameterized.expand(REPLAY_BUFFER_GYM_TESTS_2)
|
||||
@pytest.mark.serial
|
||||
@unittest.skipIf(not torch.cuda.is_available(), "CUDA not available")
|
||||
def test_replay_buffer_gym_gpu_2(self, name: str, config_path: str):
|
||||
self._test_replay_buffer_gym_gpu(name, config_path)
|
||||
|
||||
def _test_replay_buffer_gym_gpu(self, name: str, config_path: str):
|
||||
logger.info(f"Starting {name} on GPU")
|
||||
self.run_from_config(
|
||||
run_test=run_test_replay_buffer,
|
||||
@@ -263,7 +284,12 @@ def run_test_replay_buffer(
|
||||
device=device,
|
||||
)
|
||||
data_loader = torch.utils.data.DataLoader(dataset, collate_fn=identity_collate)
|
||||
pl_trainer = pl.Trainer(max_epochs=1, gpus=int(use_gpu))
|
||||
pl_trainer = pl.Trainer(
|
||||
max_epochs=1,
|
||||
gpus=int(use_gpu),
|
||||
deterministic=True,
|
||||
default_root_dir=f"lightning_log_{str(uuid.uuid4())}",
|
||||
)
|
||||
# Note: the fit() function below also evaluates the agent along the way
|
||||
# and adds the new transitions to the replay buffer, so it is training
|
||||
# on incrementally larger and larger buffers.
|
||||
@@ -311,7 +337,12 @@ def run_test_online_episode(
|
||||
agent = Agent.create_for_env(env, policy, device=device)
|
||||
|
||||
if isinstance(trainer, pl.LightningModule):
|
||||
pl_trainer = pl.Trainer(max_epochs=1, gpus=int(use_gpu), deterministic=True)
|
||||
pl_trainer = pl.Trainer(
|
||||
max_epochs=1,
|
||||
gpus=int(use_gpu),
|
||||
deterministic=True,
|
||||
default_root_dir=f"lightning_log_{str(uuid.uuid4())}",
|
||||
)
|
||||
dataset = EpisodicDataset(
|
||||
env=env, agent=agent, num_episodes=num_train_episodes, seed=SEED
|
||||
)
|
||||
|
||||
@@ -64,7 +64,7 @@ def replay_buffer_to_pre_timeline_df(
|
||||
"ds": [DEFAULT_DS for _ in range(n)],
|
||||
"state_features": _dense_to_sparse(batch.state),
|
||||
"action": action,
|
||||
"mdp_id": batch.mdp_id.tolist(),
|
||||
"mdp_id": list(map(str, batch.mdp_id.flatten().tolist())),
|
||||
"sequence_number": sequence_number,
|
||||
"action_probability": action_probability,
|
||||
"reward": reward,
|
||||
|
||||
@@ -121,7 +121,8 @@ def train_and_eval(trainer, data, num_eval_batches=100, max_epochs=1):
|
||||
train_dataloader = DataLoader(data[:-num_eval_batches], collate_fn=lambda x: x[0])
|
||||
eval_data = data[-num_eval_batches:]
|
||||
|
||||
pl_trainer = pl.Trainer(max_epochs=max_epochs)
|
||||
# disable logging in tests
|
||||
pl_trainer = pl.Trainer(max_epochs=max_epochs, logger=False)
|
||||
pl_trainer.fit(trainer, train_dataloader)
|
||||
|
||||
total_loss = 0
|
||||
|
||||
@@ -262,8 +262,7 @@ def train_workflow(
|
||||
|
||||
output_paths = {}
|
||||
for module_name, serving_module in model_manager.build_serving_modules().items():
|
||||
# TODO: make this a parameter
|
||||
torchscript_output_path = f"model_{round(time.time())}.torchscript"
|
||||
torchscript_output_path = f"{model_manager.__class__.__name__}_{module_name}_{round(time.time())}.torchscript"
|
||||
torch.jit.save(serving_module, torchscript_output_path)
|
||||
logger.info(f"Saved {module_name} to {torchscript_output_path}")
|
||||
output_paths[module_name] = torchscript_output_path
|
||||
|
||||
@@ -1 +1 @@
|
||||
python>=3.7
|
||||
python>=3.8
|
||||
|
||||
@@ -10,7 +10,7 @@ license = BSD 3-Clause License
|
||||
|
||||
[options]
|
||||
packages = find:
|
||||
python_requires = >=3.7
|
||||
python_requires = >=3.8
|
||||
install_requires =
|
||||
click>=7.0
|
||||
# ~=1.2.0 for compatibility with gym
|
||||
@@ -25,7 +25,7 @@ install_requires =
|
||||
tqdm>=4.46.0
|
||||
petastorm>=0.9.0
|
||||
parameterized>=0.7.4
|
||||
pyspark==2.4.6
|
||||
pyspark==3.1.1
|
||||
pytorch-lightning==1.1.5
|
||||
ruamel.yaml>=0.15.99
|
||||
scipy>=1.3.1
|
||||
|
||||
@@ -3,13 +3,16 @@
|
||||
# test suite on all supported python versions. To use it, "pip install tox"
|
||||
# and then run "tox" from this directory.
|
||||
|
||||
[tox]
|
||||
envlist = py37
|
||||
# This post discusses how to specify patterns for testing specific tests
|
||||
# https://stackoverflow.com/questions/36456920/is-there-a-way-to-specify-which-pytest-tests-to-run-from-a-file
|
||||
|
||||
# install CUDA 10.1 Torch
|
||||
[tox]
|
||||
envlist = py38
|
||||
|
||||
# install CUDA 10.2 Torch
|
||||
[ubuntu_gpu]
|
||||
install_command =
|
||||
pip install --pre -f https://download.pytorch.org/whl/nightly/cu101/torch_nightly.html {opts} {packages}
|
||||
pip install --pre -f https://download.pytorch.org/whl/nightly/cu102/torch_nightly.html {opts} {packages} --progress-bar off
|
||||
|
||||
[pytest]
|
||||
addopts = --verbose -d --tx popen --cov=reagent --cov-report=xml --cov-append --junitxml={envlogdir}/junit-{envname}.xml
|
||||
@@ -25,25 +28,72 @@ extras =
|
||||
gym
|
||||
test
|
||||
install_command =
|
||||
pip install --pre -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html {opts} {packages}
|
||||
pip install --pre -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html {opts} {packages} --progress-bar off
|
||||
commands =
|
||||
pytest -n 4 -m "(not serial) and (not seq2slate_long)"
|
||||
pytest -n2 -m "(not serial) and (not seq2slate_long)"
|
||||
pytest -n0 -m "serial"
|
||||
|
||||
[testenv:circleci_unittest]
|
||||
[testenv:circleci_misc_unittest]
|
||||
install_command = {[ubuntu_gpu]install_command}
|
||||
commands =
|
||||
pytest reagent/test -n auto -m "(not serial) and (not seq2slate_long)"
|
||||
pytest reagent/test -n0 -m "serial"
|
||||
pytest reagent/test -n2 -m "not serial" --ignore=reagent/test/ranking/ --ignore=reagent/test/training/ --ignore=reagent/test/prediction/ --ignore=reagent/test/world_model/
|
||||
pytest reagent/test -n0 -m "serial" --ignore=reagent/test/ranking/ --ignore=reagent/test/training/ --ignore=reagent/test/prediction/ --ignore=reagent/test/world_model/
|
||||
|
||||
[testenv:circleci_gym_unittest]
|
||||
|
||||
[testenv:circleci_gym_replay_buffer_1_cpu_unittest]
|
||||
commands =
|
||||
pytest reagent/gym/tests -n2 -m "not serial" -k "test_replay_buffer_gym_cpu_1"
|
||||
|
||||
|
||||
[testenv:circleci_gym_replay_buffer_2_cpu_unittest]
|
||||
commands =
|
||||
pytest reagent/gym/tests -n2 -m "not serial" -k "test_replay_buffer_gym_cpu_2"
|
||||
|
||||
|
||||
# all cpu tests in reagent/gym/tests except test_replay_buffer_gym_cpu_x
|
||||
[testenv:circleci_gym_cpu_unittest]
|
||||
commands =
|
||||
pytest reagent/gym/tests -n2 -m "not serial" -k "not test_replay_buffer_gym_cpu"
|
||||
|
||||
|
||||
[testenv:circleci_gym_replay_buffer_1_gpu_unittest]
|
||||
install_command = {[ubuntu_gpu]install_command}
|
||||
commands =
|
||||
pytest reagent/gym/tests -n2 -m "(not serial) and (not seq2slate_long)"
|
||||
pytest reagent/gym/tests -n0 -m "serial"
|
||||
pytest reagent/gym/tests -n0 -m "serial" -k "test_replay_buffer_gym_gpu_1"
|
||||
|
||||
|
||||
[testenv:circleci_seq2slate_unittest]
|
||||
[testenv:circleci_gym_replay_buffer_2_gpu_unittest]
|
||||
install_command = {[ubuntu_gpu]install_command}
|
||||
commands =
|
||||
pytest reagent/test -n0 -m "seq2slate_long"
|
||||
pytest reagent/gym/tests -n0 -m "serial" -k "test_replay_buffer_gym_gpu_2"
|
||||
|
||||
|
||||
# all gpu tests in reagent/gym/tests except test_replay_buffer_gym_gpu_x
|
||||
[testenv:circleci_gym_gpu_unittest]
|
||||
install_command = {[ubuntu_gpu]install_command}
|
||||
commands =
|
||||
pytest reagent/gym/tests -n0 -m "serial" -k "not test_replay_buffer_gym_gpu"
|
||||
|
||||
|
||||
[testenv:circleci_ranking_unittest]
|
||||
install_command = {[ubuntu_gpu]install_command}
|
||||
commands =
|
||||
pytest reagent/test/ranking -n2
|
||||
|
||||
|
||||
[testenv:circleci_training_unittest]
|
||||
install_command = {[ubuntu_gpu]install_command}
|
||||
commands =
|
||||
pytest reagent/test/training -n2
|
||||
|
||||
|
||||
[testenv:circleci_prediction_unittest]
|
||||
install_command = {[ubuntu_gpu]install_command}
|
||||
commands =
|
||||
pytest reagent/test/prediction -n2
|
||||
|
||||
|
||||
[testenv:circleci_world_model_unittest]
|
||||
install_command = {[ubuntu_gpu]install_command}
|
||||
commands =
|
||||
pytest reagent/test/world_model -n2
|
||||
|
||||
Reference in New Issue
Block a user