d4rl_adroit_pen

  • Descrizione :

D4RL è un punto di riferimento open source per l'apprendimento per rinforzo offline. Fornisce ambienti e set di dati standardizzati per algoritmi di training e benchmarking.

I set di dati seguono il formato RLDS per rappresentare passaggi ed episodi.

@misc{fu2020d4rl,
    title
={D4RL: Datasets for Deep Data-Driven Reinforcement Learning},
    author
={Justin Fu and Aviral Kumar and Ofir Nachum and George Tucker and Sergey Levine},
    year
={2020},
    eprint
={2004.07219},
    archivePrefix
={arXiv},
    primaryClass
={cs.LG}
}

d4rl_adroit_pen/v0-human (configurazione predefinita)

Diviso Esempi
'train' 50
  • Struttura delle caratteristiche :
FeaturesDict({
   
'steps': Dataset({
       
'action': Tensor(shape=(24,), dtype=float32),
       
'discount': float32,
       
'infos': FeaturesDict({
           
'qpos': Tensor(shape=(30,), dtype=float32),
           
'qvel': Tensor(shape=(30,), dtype=float32),
       
}),
       
'is_first': bool,
       
'is_last': bool,
       
'is_terminal': bool,
       
'observation': Tensor(shape=(45,), dtype=float32),
       
'reward': float32,
   
}),
})
  • Documentazione delle funzionalità :
Caratteristica Classe Forma Tipo D Descrizione
CaratteristicheDict
passi Set di dati
passi/azione Tensore (24,) float32
passi/sconto Tensore float32
passaggi/informazioni CaratteristicheDict
passi/informazioni/qpos Tensore (30,) float32
passaggi/informazioni/qvel Tensore (30,) float32
passi/è_primo Tensore bool
passi/è_ultimo Tensore bool
passi/è_terminale Tensore bool
passi/osservazione Tensore (45,) float32
passi/ricompensa Tensore float32

d4rl_adroit_pen/v0-cloned

Diviso Esempi
'train' 5.023
  • Struttura delle caratteristiche :
FeaturesDict({
   
'steps': Dataset({
       
'action': Tensor(shape=(24,), dtype=float32),
       
'discount': float64,
       
'infos': FeaturesDict({
           
'qpos': Tensor(shape=(30,), dtype=float64),
           
'qvel': Tensor(shape=(30,), dtype=float64),
       
}),
       
'is_first': bool,
       
'is_last': bool,
       
'is_terminal': bool,
       
'observation': Tensor(shape=(45,), dtype=float64),
       
'reward': float64,
   
}),
})
  • Documentazione delle funzionalità :
Caratteristica Classe Forma Tipo D Descrizione
CaratteristicheDict
passi Set di dati
passi/azione Tensore (24,) float32
passi/sconto Tensore float64
passaggi/informazioni CaratteristicheDict
passi/informazioni/qpos Tensore (30,) float64
passaggi/informazioni/qvel Tensore (30,) float64
passi/è_primo Tensore bool
passi/è_ultimo Tensore bool
passi/è_terminale Tensore bool
passi/osservazione Tensore (45,) float64
passi/ricompensa Tensore float64

d4rl_adroit_pen/v0-expert

Diviso Esempi
'train' 5.000
  • Struttura delle caratteristiche :
FeaturesDict({
   
'steps': Dataset({
       
'action': Tensor(shape=(24,), dtype=float32),
       
'discount': float32,
       
'infos': FeaturesDict({
           
'action_logstd': Tensor(shape=(24,), dtype=float32),
           
'action_mean': Tensor(shape=(24,), dtype=float32),
           
'qpos': Tensor(shape=(30,), dtype=float32),
           
'qvel': Tensor(shape=(30,), dtype=float32),
       
}),
       
'is_first': bool,
       
'is_last': bool,
       
'is_terminal': bool,
       
'observation': Tensor(shape=(45,), dtype=float32),
       
'reward': float32,
   
}),
})
  • Documentazione delle funzionalità :
Caratteristica Classe Forma Tipo D Descrizione
CaratteristicheDict
passi Set di dati
passi/azione Tensore (24,) float32
passi/sconto Tensore float32
passaggi/informazioni CaratteristicheDict
passaggi/info/action_logstd Tensore (24,) float32
passi/info/azione_mean Tensore (24,) float32
passi/informazioni/qpos Tensore (30,) float32
passaggi/informazioni/qvel Tensore (30,) float32
passi/è_primo Tensore bool
passi/è_ultimo Tensore bool
passi/è_terminale Tensore bool
passi/osservazione Tensore (45,) float32
passi/ricompensa Tensore float32

d4rl_adroit_pen/v1-human

Diviso Esempi
'train' 25
  • Struttura delle caratteristiche :
FeaturesDict({
   
'steps': Dataset({
       
'action': Tensor(shape=(24,), dtype=float32),
       
'discount': float32,
       
'infos': FeaturesDict({
           
'desired_orien': Tensor(shape=(4,), dtype=float32),
           
'qpos': Tensor(shape=(30,), dtype=float32),
           
'qvel': Tensor(shape=(30,), dtype=float32),
       
}),
       
'is_first': bool,
       
'is_last': bool,
       
'is_terminal': bool,
       
'observation': Tensor(shape=(45,), dtype=float32),
       
'reward': float32,
   
}),
})
  • Documentazione delle funzionalità :
Caratteristica Classe Forma Tipo D Descrizione
CaratteristicheDict
passi Set di dati
passi/azione Tensore (24,) float32
passi/sconto Tensore float32
passaggi/informazioni CaratteristicheDict
passaggi/info/desired_orien Tensore (4,) float32
passi/informazioni/qpos Tensore (30,) float32
passaggi/informazioni/qvel Tensore (30,) float32
passi/è_primo Tensore bool
passi/è_ultimo Tensore bool
passi/è_terminale Tensore bool
passi/osservazione Tensore (45,) float32
passi/ricompensa Tensore float32

d4rl_adroit_pen/v1-cloned

Diviso Esempi
'train' 3.755
  • Struttura delle caratteristiche :
FeaturesDict({
   
'algorithm': string,
   
'policy': FeaturesDict({
       
'fc0': FeaturesDict({
           
'bias': Tensor(shape=(256,), dtype=float32),
           
'weight': Tensor(shape=(45, 256), dtype=float32),
       
}),
       
'fc1': FeaturesDict({
           
'bias': Tensor(shape=(256,), dtype=float32),
           
'weight': Tensor(shape=(256, 256), dtype=float32),
       
}),
       
'last_fc': FeaturesDict({
           
'bias': Tensor(shape=(24,), dtype=float32),
           
'weight': Tensor(shape=(256, 24), dtype=float32),
       
}),
       
'nonlinearity': string,
       
'output_distribution': string,
   
}),
   
'steps': Dataset({
       
'action': Tensor(shape=(24,), dtype=float32),
       
'discount': float32,
       
'infos': FeaturesDict({
           
'desired_orien': Tensor(shape=(4,), dtype=float32),
           
'qpos': Tensor(shape=(30,), dtype=float32),
           
'qvel': Tensor(shape=(30,), dtype=float32),
       
}),
       
'is_first': bool,
       
'is_last': bool,
       
'is_terminal': bool,
       
'observation': Tensor(shape=(45,), dtype=float32),
       
'reward': float32,
   
}),
})
  • Documentazione delle funzionalità :
Caratteristica Classe Forma Tipo D Descrizione
CaratteristicheDict
algoritmo Tensore corda
politica CaratteristicheDict
polizza/fc0 CaratteristicheDict
policy/fc0/bias Tensore (256,) float32
politica/fc0/peso Tensore (45, 256) float32
politica/fc1 CaratteristicheDict
policy/fc1/bias Tensore (256,) float32
politica/fc1/peso Tensore (256, 256) float32
politica/last_fc CaratteristicheDict
policy/last_fc/bias Tensore (24,) float32
politica/last_fc/peso Tensore (256, 24) float32
politica/non linearità Tensore corda
politica/distribuzione_output Tensore corda
passi Set di dati
passi/azione Tensore (24,) float32
passi/sconto Tensore float32
passaggi/informazioni CaratteristicheDict
passaggi/info/desired_orien Tensore (4,) float32
passi/informazioni/qpos Tensore (30,) float32
passaggi/informazioni/qvel Tensore (30,) float32
passi/è_primo Tensore bool
passi/è_ultimo Tensore bool
passi/è_terminale Tensore bool
passi/osservazione Tensore (45,) float32
passi/ricompensa Tensore float32

d4rl_adroit_pen/v1-expert

  • Dimensione download : 249.90 MiB

  • Dimensione del set di dati : 548.47 MiB

  • Memorizzazione nella cache automatica ( documentazione ): No

  • Divide :

Diviso Esempi
'train' 5.000
  • Struttura delle caratteristiche :
FeaturesDict({
   
'algorithm': string,
   
'policy': FeaturesDict({
       
'fc0': FeaturesDict({
           
'bias': Tensor(shape=(64,), dtype=float32),
           
'weight': Tensor(shape=(64, 45), dtype=float32),
       
}),
       
'fc1': FeaturesDict({
           
'bias': Tensor(shape=(64,), dtype=float32),
           
'weight': Tensor(shape=(64, 64), dtype=float32),
       
}),
       
'last_fc': FeaturesDict({
           
'bias': Tensor(shape=(24,), dtype=float32),
           
'weight': Tensor(shape=(24, 64), dtype=float32),
       
}),
       
'last_fc_log_std': FeaturesDict({
           
'bias': Tensor(shape=(24,), dtype=float32),
           
'weight': Tensor(shape=(24, 64), dtype=float32),
       
}),
       
'nonlinearity': string,
       
'output_distribution': string,
   
}),
   
'steps': Dataset({
       
'action': Tensor(shape=(24,), dtype=float32),
       
'discount': float32,
       
'infos': FeaturesDict({
           
'action_log_std': Tensor(shape=(24,), dtype=float32),
           
'action_mean': Tensor(shape=(24,), dtype=float32),
           
'desired_orien': Tensor(shape=(4,), dtype=float32),
           
'qpos': Tensor(shape=(30,), dtype=float32),
           
'qvel': Tensor(shape=(30,), dtype=float32),
       
}),
       
'is_first': bool,
       
'is_last': bool,
       
'is_terminal': bool,
       
'observation': Tensor(shape=(45,), dtype=float32),
       
'reward': float32,
   
}),
})
  • Documentazione delle funzionalità :
Caratteristica Classe Forma Tipo D Descrizione
CaratteristicheDict
algoritmo Tensore corda
politica CaratteristicheDict
polizza/fc0 CaratteristicheDict
policy/fc0/bias Tensore (64,) float32
politica/fc0/peso Tensore (64, 45) float32
politica/fc1 CaratteristicheDict
policy/fc1/bias Tensore (64,) float32
politica/fc1/peso Tensore (64, 64) float32
politica/last_fc CaratteristicheDict
policy/last_fc/bias Tensore (24,) float32
politica/last_fc/peso Tensore (24, 64) float32
policy/last_fc_log_std CaratteristicheDict
policy/last_fc_log_std/bias Tensore (24,) float32
policy/last_fc_log_std/weight Tensore (24, 64) float32
politica/non linearità Tensore corda
politica/distribuzione_output Tensore corda
passi Set di dati
passi/azione Tensore (24,) float32
passi/sconto Tensore float32
passaggi/informazioni CaratteristicheDict
passaggi/info/action_log_std Tensore (24,) float32
passi/info/azione_mean Tensore (24,) float32
passaggi/info/desired_orien Tensore (4,) float32
passi/informazioni/qpos Tensore (30,) float32
passaggi/informazioni/qvel Tensore (30,) float32
passi/è_primo Tensore bool
passi/è_ultimo Tensore bool
passi/è_terminale Tensore bool
passi/osservazione Tensore (45,) float32
passi/ricompensa Tensore float32