# print all parameters again (using the alias) aliasAlias = myData[...].aliasAlias print('initial aliasAlias\n', aliasAlias, '\n\n') # update the parameters via the alias aliasAlias[1] = [5, 5, 3, 5, 5] # store the updated parameters myData[1].aliasAlias = aliasAlias[1] # print all parameters one more time print('updated parameters\n', myData[...].parameters, '\n\n') # print all parameters again (using the alias) print('updated context\n', myData[...].contexts, '\n\n') # print all parameters again (using the alias) print('updated aliasAlias\n', myData[...].aliasAlias, '\n\n') # test vertical alignment for alias print('vertical Alias\n', myData[...].parametersAliasVertical, '\n\n') # test vertical alignment for alias myData[...].parametersAliasVertical = myData[...].parametersAliasVertical * 2 print('vertical Alias after resetting values\n', myData[...].parametersAliasVertical, '\n\n') print('Original Parameters after resetting values\n', myData[...].parameters, '\n\n') print('Num Dim Vertical Parameters\n', dataManager.getNumDimensions('parametersAliasVertical'), '\n\n')
print('Timing:', t0 - t1) t0 = time.time() print('Next states:', myData[...].nextStates) t1 = time.time() print('Timing:', t0 - t1) t0 = time.time() print('Previous states:', myData[..., slice(0, 10)].lastStates) t1 = time.time() print('Timing:', t0 - t1) t0 = time.time() print('All states:', myData[...].allStates) t1 = time.time() print('Timing:', t0 - t1) t0 = time.time() print('History states:', myData[...].historyStates.shape) t1 = time.time() print('Timing:', t0 - t1) t0 = time.time() # get history states for each episode for i in range(0, 100): print('History states ', i, ' shape:', myData[i].historyStates.shape) t1 = time.time() print('Timing:', t0 - t1) print('History Dimensions:', dataManager.getNumDimensions('historyStates'))
intra_op_parallelism_threads=num_cpu) session = tf.Session(config=tf_config) session.__enter__() dataManager = DataManager('data') dataManager.addDataEntry('states', 4) dataManager.addDataEntry('actions', 6) settings = getDefaultSettings() gaussian = LinearFullGaussian(dataManager, ['states'], ['actions']) data = dataManager.createDataObject([10000]) data[...].states = np.random.normal(0, 1, data[...].states.shape) A = np.random.normal(0, 1, (dataManager.getNumDimensions('actions'), dataManager.getNumDimensions('actions'))) covMat = A.dot(A.transpose()) gaussian.param_stdmat = np.linalg.cholesky(covMat) gaussian.param_final_w = np.random.normal(0, 1, gaussian.param_final_w.shape) gaussian.param_final_b = np.random.normal(0, 1, gaussian.param_final_b.shape) # Sample from Gaussian and write back in data data[...] >> gaussian >> data # Compute log likelihood gaussianLearned = LinearFullGaussian(dataManager, ['states'], ['actions']) learner = LinearGaussianMLLearner(dataManager, gaussianLearned) data[...] >> learner