notebook.community

Edit and run



In [1]:

    
# July 31 2017
# This file is for experiments with student2 4 skills and training trajectory length 7
# to check different architectures and whether they can learn good models

%load_ext autoreload
%autoreload 2
%matplotlib inline

import numpy as np
import scipy as sp
import six
from matplotlib.pyplot import *



In [10]:

    
'''
Default Double LSTM
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_model_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/stats-runA.npz')

vloss = data11['vloss']
#scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00126
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

xlim(10, 15)
ylim(0.0012,0.0014)

# looks like epoch 11 is good for 0.00001
# but might as well just make it 12 to be consistent with the others









    



(10, 21)
last 0.00125380288737 threshold 0.00126






    Out[10]:





(0.0012, 0.0014)



In [14]:

    
'''
Single LSTM
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_modelsimple_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/stats-runA.npz')

vloss = data11['vloss']
#scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00001
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

#xlim(10, 15)
#ylim(0.00,0.00002)

# looks like epoch 12 is good for 0.00001









    



(10, 21)
last 7.97343786871e-08 threshold 1e-05






    Out[14]:





[<matplotlib.lines.Line2D at 0x7f9001abcf90>]



In [25]:

    
'''
Double GRU
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_modelgru_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/stats-runA.npz')

vloss = data11['vloss']
#scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00009
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

xlim(10, 15)
ylim(0.00,0.0001)

# looks like epoch 12 is good for 0.00001









    



(10, 21)
last 8.66475158124e-05 threshold 9e-05






    Out[25]:





(0.0, 0.0001)



In [7]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Default Double LSTM
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_model_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_model_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')









    



vloss shape (50, 13)
scores shape (50,)
[ 1.          0.125       0.          0.30208333  0.          0.          0.
  0.          0.          0.13541667  0.          0.          0.0625      0.
  0.          0.94791667  0.          0.          0.          0.          0.
  1.          1.          1.          0.          0.          0.41666667
  0.          0.86458333  0.61458333  1.          0.          0.          0.
  0.          0.33333333  0.          0.01041667  0.26041667  0.          1.
  0.          0.96875     0.46875     0.          0.          0.375       0.
  0.          0.        ]






    Out[7]:





[<matplotlib.lines.Line2D at 0x7f356fce2f90>]



In [8]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Single LSTM
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_modelsimple_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_modelsimple_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')









    



vloss shape (50, 13)
scores shape (50,)
[ 0.97916667  0.          0.          0.          0.          0.          0.
  0.          0.04166667  0.          0.          0.          0.
  0.54166667  0.58333333  0.          0.          0.53125     0.51041667
  0.96875     0.          0.          0.          0.          0.          0.
  0.          1.          0.1875      0.          0.          0.          1.
  0.          0.          0.          0.          0.57291667  0.          0.
  0.          1.          0.          0.          0.          1.          0.
  0.          0.92708333  0.        ]






    Out[8]:





[<matplotlib.lines.Line2D at 0x7f356db77590>]



In [9]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Double GRU
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_modelgru_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_modelgru_small-dropout10-shuffle0-data-test2-n100000-l7-random.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')









    



vloss shape (50, 13)
scores shape (50,)
[ 1.          0.98958333  0.01041667  0.          0.          0.          1.
  0.          0.          0.          0.98958333  0.          0.97916667
  0.          0.91666667  0.          0.01041667  0.          0.          0.
  0.          0.          0.97916667  0.          0.          0.          1.
  0.          0.54166667  0.          0.          0.10416667  0.          0.
  0.          0.          0.          1.          0.          0.
  0.19791667  0.          0.16666667  0.          0.          0.01041667
  0.          0.01041667  0.          0.        ]






    Out[9]:





[<matplotlib.lines.Line2D at 0x7f356dc77e50>]

Now switched to using egreedy 0.3 instead of random exploration



In [9]:

    
'''
Default Double LSTM
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_model_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runA.npz')

vloss = data11['vloss']
#scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00146
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

xlim(5, 10)
ylim(0.0014,0.0015)

# looks like epoch 9 is good for 0.00001









    



(10, 21)
last 0.00145415554735 threshold 0.00146






    Out[9]:





(0.0014, 0.0015)



In [13]:

    
'''
Single LSTM
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_modelsimple_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runA.npz')

vloss = data11['vloss']
#scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00001
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

xlim(5, 10)
ylim(0.00,0.00002)

# looks like epoch 8 is good for 0.00001









    



(10, 21)
last 9.48756607707e-09 threshold 1e-05






    Out[13]:





(0.0, 2e-05)



In [5]:

    
'''
Double GRU.
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_modelgru_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runA.npz')

vloss = data11['vloss']
#scores = data21['scores'][:,0]
#initialq = data51['qvals'][:,0]
#opts = data61['opts']
#qfuncs = data61['qs'][:,0,:,:]

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00001
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

#xlim(5, 10)
#ylim(0.00,0.00002)

# looks like epoch 6 is good for 0.00001









    



(10, 21)
last 1.97803311244e-07 threshold 1e-05






    Out[5]:





[<matplotlib.lines.Line2D at 0x7f2f4aa3acd0>]



In [2]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Behavior policy is egreedy0.3
Default Double LSTM
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_model_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_model_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')

# 4 good models, 2 more kind of good models









    



vloss shape (50, 10)
scores shape (50,)
[ 0.          0.90625     0.          1.          0.          0.          0.
  0.          0.          0.01041667  0.          0.          0.125       0.
  0.03125     0.          0.          0.          0.          0.          0.
  0.          0.          0.67708333  0.84375     0.          1.          0.
  0.          0.          0.          0.          0.05208333  0.05208333
  0.0625      0.          0.          0.85416667  0.          0.          0.
  0.29166667  0.9375      0.09375     0.          0.          0.          0.
  0.10416667  0.        ]






    Out[2]:





[<matplotlib.lines.Line2D at 0x7f2f4d22f510>]



In [3]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Behavior policy is egreedy0.3
Single LSTM
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_modelsimple_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_modelsimple_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')

# 6 good models, 2 more kinda good models









    



vloss shape (50, 9)
scores shape (50,)
[ 0.0625      0.          0.          0.          0.          0.89583333
  0.          0.          0.          0.84375     0.          0.          0.
  0.          0.15625     0.          1.          0.10416667  0.          0.
  0.          1.          0.          0.95833333  0.          0.          0.
  0.          0.          0.          0.29166667  0.          0.          0.
  0.76041667  0.          0.          0.          0.39583333  0.
  0.91666667  0.          0.          0.91666667  0.          1.          0.
  0.          0.          0.        ]






    Out[3]:





[<matplotlib.lines.Line2D at 0x7f2f4acf1890>]



In [4]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Behavior policy is egreedy0.3
Double GRU
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_modelgru_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_modelgru_small-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')

# 4 good models









    



vloss shape (50, 7)
scores shape (50,)
[ 0.          0.          0.          0.55208333  0.          1.
  0.39583333  0.67708333  0.36458333  0.          0.6875      0.59375     0.
  0.76041667  0.73958333  0.92708333  1.          0.15625     0.02083333
  0.30208333  0.          0.04166667  0.          0.          0.          0.
  0.64583333  0.          0.          0.01041667  0.          0.          0.5
  0.02083333  0.          0.          0.91666667  0.22916667  0.01041667
  0.63541667  0.          0.          0.          0.01041667  0.          0.
  0.          0.          0.          0.        ]






    Out[4]:





[<matplotlib.lines.Line2D at 0x7f2f4acbcb10>]

Trying out mid model (10 hidden units) with egreedy 0.30



In [7]:

    
'''
Default Double LSTM
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_model_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runA.npz')

vloss = data11['vloss']

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00146
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

xlim(5, 10)
ylim(0.0014,0.0015)

# looks like epoch 6 is good for 0.00001









    



(10, 21)
last 0.00145032981119 threshold 0.00146






    Out[7]:





(0.0014, 0.0015)



In [10]:

    
'''
Single LSTM
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_modelsimple_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runA.npz')

vloss = data11['vloss']

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00001
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

xlim(0, 6)
ylim(0.00,0.0001)

# looks like epoch 5 is good for 0.00001









    



(10, 21)
last 1.47997240061e-09 threshold 1e-05






    Out[10]:





(0.0, 0.0001)



In [11]:

    
'''
Double GRU
Trying to find a training epoch to stop training.
'''
data11 = np.load('experiments/test2_modelgru_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runA.npz')

vloss = data11['vloss']

six.print_(vloss.shape)
figure()
x_avg = np.mean(vloss,axis=0)
x_serr = np.std(vloss,axis=0) / vloss.shape[0] * 1.98
plot(x_avg, color='#0000ff')
plot(x_avg + x_serr, color='#ddddff')
plot(x_avg - x_serr, color='#ddddff')

x_end = np.mean(x_avg[-2:])
x_thres = 0.00001
six.print_('last {} threshold {}'.format(x_end, x_thres))
plot([0,20],[x_end,x_end],color='#ff0000')
plot([0,20],[x_thres,x_thres],color='#ff0000')

xlim(0, 6)
ylim(0.00,0.0001)

# looks like epoch 4 is good for 0.00001









    



(10, 21)
last 1.16617886051e-09 threshold 1e-05






    Out[11]:





(0.0, 0.0001)



In [2]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Behavior policy is egreedy0.3
Default Double LSTM
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_model_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_model_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')

# 5 models >90%









    



vloss shape (50, 7)
scores shape (50,)
[ 0.          0.          0.90625     0.          0.          0.          0.
  0.          0.          1.          0.78125     0.5625      0.          0.
  1.          0.          0.          0.          0.          0.01041667
  0.          0.          0.71875     0.          0.          0.          0.
  0.51041667  0.          0.          0.          0.          0.          0.
  0.          0.          1.          0.          0.          0.          0.
  0.          0.125       0.          0.875       0.          0.
  0.94791667  0.          0.        ]






    Out[2]:





[<matplotlib.lines.Line2D at 0x7f33b8abe510>]



In [3]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Behavior policy is egreedy0.3
Single LSTM
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_modelsimple_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_modelsimple_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')

# 6 models >90%









    



vloss shape (50, 6)
scores shape (50,)
[ 0.          0.13541667  0.          0.          0.          0.08333333
  0.32291667  0.0625      0.          0.97916667  0.          0.          0.
  0.875       0.          0.          0.01041667  0.          0.          0.
  0.          0.          0.          0.          0.17708333  0.
  0.94791667  0.          1.          0.          0.86458333  0.          0.
  0.          0.          0.          0.          0.          0.          0.
  0.03125     0.          0.          0.          1.          0.
  0.98958333  0.          0.96875     0.        ]






    Out[3]:





[<matplotlib.lines.Line2D at 0x7f33b87ea290>]



In [4]:

    
'''
Analyzing results of student2 with 4 skills, with training length 7 and testing length 6.
Behavior policy is egreedy0.3
Double GRU
Looking at correlation between training AUC and actual performance
'''
data11 = np.load('experiments/test2_modelgru_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/stats-runB.npz')
data21 = np.load('experiments/test2_modelgru_mid-dropout10-shuffle0-data-test2-n100000-l7-egreedy0.30.pickle/mcts-rtype2-rollouts3000-trajectories100-real1-runB.npz')

vloss = data11['vloss']
scores = data21['scores'][:,0]

six.print_('vloss shape {}'.format(vloss.shape))
#six.print_(vloss)
six.print_('scores shape {}'.format(scores.shape))
six.print_(scores)

xs = np.sum(vloss,axis=1)
ys = scores
title('Training Loss AUC versus Actual Performance')
xlabel('Training Loss AUC')
ylabel('Posttest Score')
plot(xs,ys,'.')

# 3 models >90%









    



vloss shape (50, 5)
scores shape (50,)
[ 0.          0.10416667  0.          0.82291667  0.          0.66666667
  0.69791667  0.90625     0.          0.          0.          0.          1.
  0.          0.          0.78125     0.          0.          0.52083333
  0.9375      0.          0.          0.35416667  0.          0.          0.
  0.67708333  0.          0.89583333  0.          0.          0.          0.
  0.          0.          0.          0.64583333  0.          0.85416667
  0.          0.84375     0.625       0.          0.          0.          0.
  0.76041667  0.          0.          0.        ]






    Out[4]:





[<matplotlib.lines.Line2D at 0x7f33b65c8090>]



In [ ]: