File size: 7,032 Bytes
079c32c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
apiVersion: diengine.opendilab.org/v1alpha1
kind: DIJob
metadata:
  name: cartpole-dqn
  labels: 
    run-dijob-type: test
spec:
  group: xxx
  priorityClassName: ""
  cleanPodPolicy: "Running"
  volumes:
  - name: cache-volume
    emptyDir:
      medium: Memory
      sizeLimit: 128Mi
  - name: work-dir
    hostPath:
      path: /data/di-engine
  coordinator:
    template:
      spec:
        containers:
        - name: di-container
          image: diorchestrator/ding:v0.1.1
          imagePullPolicy: IfNotPresent
          env:
          - name: PYTHONUNBUFFERED
            value: "1"
          command: ["/bin/bash", "-c",]
          args:
          - |
            cat <<EOF > cartpole_dqn_config_k8s.py
            from easydict import EasyDict

            cartpole_dqn_config = dict(
                exp_name='cartpole_dqn',
                env=dict(
                    collector_env_num=8,
                    collector_episode_num=2,
                    evaluator_env_num=5,
                    evaluator_episode_num=1,
                    stop_value=195,
                ),
                policy=dict(
                    cuda=False,
                    model=dict(
                        obs_shape=4,
                        action_shape=2,
                        encoder_hidden_size_list=[128, 128, 64],
                        dueling=True,
                    ),
                    nstep=3,
                    discount_factor=0.97,
                    learn=dict(
                        batch_size=32,
                        learning_rate=0.001,
                        learner=dict(
                            learner_num=1,
                            send_policy_freq=1,
                        ),
                    ),
                    collect=dict(
                        n_sample=16,
                        collector=dict(
                            collector_num=2,
                            update_policy_second=3,
                        ),
                    ),
                    eval=dict(evaluator=dict(eval_freq=50, )),
                    other=dict(
                        eps=dict(
                            type='exp',
                            start=0.95,
                            end=0.1,
                            decay=100000,
                        ),
                        replay_buffer=dict(
                            replay_buffer_size=100000,
                            enable_track_used_data=False,
                        ),
                        commander=dict(
                            collector_task_space=2,
                            learner_task_space=1,
                            eval_interval=5,
                        ),
                    ),
                ),
            )
            cartpole_dqn_config = EasyDict(cartpole_dqn_config)
            main_config = cartpole_dqn_config

            cartpole_dqn_create_config = dict(
                env=dict(
                    type='cartpole',
                    import_names=['dizoo.classic_control.cartpole.envs.cartpole_env'],
                ),
                env_manager=dict(type='base'),
                policy=dict(type='dqn_command'),
                learner=dict(type='base', import_names=['ding.worker.learner.base_learner']),
                collector=dict(
                    type='zergling',
                    import_names=['ding.worker.collector.zergling_parallel_collector'],
                ),
                commander=dict(
                    type='solo',
                    import_names=['ding.worker.coordinator.solo_parallel_commander'],
                ),
                comm_learner=dict(
                    type='flask_fs',
                    import_names=['ding.worker.learner.comm.flask_fs_learner'],
                ),
                comm_collector=dict(
                    type='flask_fs',
                    import_names=['ding.worker.collector.comm.flask_fs_collector'],
                ),
            )
            cartpole_dqn_create_config = EasyDict(cartpole_dqn_create_config)
            create_config = cartpole_dqn_create_config

            cartpole_dqn_system_config = dict(
                coordinator=dict(
                    operator_server=dict(
                        system_addr='di-server.di-system:8080',
                        api_version='/v1alpha1',
                        init_replicas_request=dict(
                            collectors={
                                "replicas": 2,
                            },
                            learners={
                                "gpus": "0",
                                "replicas": 1,
                            },
                        ),
                        collector_target_num=2,
                        learner_target_num=1,
                    ),
                ),
                path_data='./{}/data'.format(main_config.exp_name),
                path_policy='./{}/policy'.format(main_config.exp_name),
                communication_mode='auto',
                learner_gpu_num=1,
            )
            cartpole_dqn_system_config = EasyDict(cartpole_dqn_system_config)
            system_config = cartpole_dqn_system_config

            if __name__ == '__main__':
                from ding.entry.parallel_entry import parallel_pipeline
                parallel_pipeline([main_config, create_config, system_config], seed=9)
            EOF

            ding -m dist --module config -P k8s -c ./cartpole_dqn_config_k8s.py -s 0;
            ding -m dist --module coordinator -c /ding/cartpole_dqn_config_k8s.py.pkl -s 0 -cdp $COORDINATOR_PORT
          ports:
          - name: di-port
            containerPort: 22270
          volumeMounts:
          - name: work-dir
            mountPath: /ding
  collector:
    template:
      spec:
        containers:
        - name: di-container
          image: diorchestrator/ding:v0.1.1
          imagePullPolicy: IfNotPresent
          env:
          - name: PYTHONUNBUFFERED
            value: "1"
          command: ["/bin/bash", "-c",]
          args:
          - |
            ding -m dist --module collector -c /ding/cartpole_dqn_config_k8s.py.pkl -s 0 -clp $COLLECTOR_PORT
          ports:
          - name: di-port
            containerPort: 22270
          volumeMounts:
          - name: work-dir
            mountPath: /ding
  learner:
    template:
      spec:
        containers:
        - name: di-container
          image: diorchestrator/ding:v0.1.1
          imagePullPolicy: IfNotPresent
          env:
          - name: PYTHONUNBUFFERED
            value: "1"
          command: ["/bin/bash", "-c",]
          args:
          - |
            ding -m dist --module spawn_learner -c /ding/cartpole_dqn_config_k8s.py.pkl -s 0 -lp $LEARNER_PORT
          ports:
          - name: di-port
            containerPort: 22270
          volumeMounts:
          - name: cache-volume
            mountPath: /dev/shm
          - name: work-dir
            mountPath: /ding