

In [1]:

    
import pymongo
from pymongo import MongoClient
import pandas as pd
import matplotlib.pyplot as plt



In [2]:

    
%matplotlib nbagg

Definitions. Fill in your own values here. If you used all default values for launching the search, only need to change exp_key



In [17]:

    
result_db_name = 'rllab'
exp_key = 'test11'
host = 'localhost'
port = 1234

Connect to database (make sure mongodb is running)



In [18]:

    
client = MongoClient(host, port)
db = client[result_db_name]
jobs = db.jobs

Retrieve and format results



In [19]:

    
df = pd.DataFrame(list(jobs.find({'exp_key':exp_key})))
param_dicts = []
score_curves = []
for i in range(df.shape[0]):
    res = df.result.iat[i]
    if res['status'] == 'ok':
        resdict = res['params']
        resdict['score'] = -df.result[i]['loss']
        resdict['result_index'] = i
        score_curves.append(res['scores']) # this only works if your function to process results returns the learning curve
        param_dicts.append(resdict)
param_df = pd.DataFrame(param_dicts)

Results table



In [20]:

    
param_df









    Out[20]:






  
    
      
      exp_name
      result_index
      score
      seed
      step_size
    
  
  
    
      0
      test11_25612_1
      0
      999.935556
      1
      0.051536
    
    
      1
      test11_25584_1
      1
      999.955588
      0
      0.035537
    
    
      2
      test11_25598_1
      2
      999.954819
      2
      0.088319
    
    
      3
      test11_26214_1
      3
      999.947394
      0
      0.071792
    
    
      4
      test11_26204_1
      4
      999.956958
      0
      0.058653

Learning curve of best parameter set



In [21]:

    
best_score_index = param_df.score.values.argmax()
plt.figure()
plt.plot(score_curves[best_score_index])
plt.xlabel('Learning algo iter')
plt.ylabel('Average return')









    














    











    Out[21]:





<matplotlib.text.Text at 0x7fb090cb2828>

Hyperopt score progression



In [22]:

    
plt.figure()
plt.plot(param_df.score.rolling(window=2).max())
plt.title('Rolling max of hyperopt score progression')
plt.xlabel('Hyperopt iter')
plt.ylabel('Score')









    














    











    Out[22]:





<matplotlib.text.Text at 0x7fb0889602b0>



In [ ]:

	exp_name	result_index	score	seed	step_size
0	test11_25612_1	0	999.935556	1	0.051536
1	test11_25584_1	1	999.955588	0	0.035537
2	test11_25598_1	2	999.954819	2	0.088319
3	test11_26214_1	3	999.947394	0	0.071792
4	test11_26204_1	4	999.956958	0	0.058653