notebook.community

Edit and run



In [1]:

    
import pylearn2.utils
import pylearn2.config
import theano
import neukrill_net.dense_dataset
import neukrill_net.utils
import numpy as np
%matplotlib inline
import matplotlib.pyplot as plt
import holoviews as hl
%load_ext holoviews.ipython
import sklearn.metrics









    



Using gpu device 3: Tesla K40c
:0: FutureWarning: IPython widgets are experimental and may change in the future.






    



Welcome to the HoloViews IPython extension! (http://ioam.github.io/holoviews/)
Available magics: %compositor, %opts, %params, %view, %%labels, %%opts, %%view






    





<matplotlib.figure.Figure at 0x7f1e51fc1b10>






    





<matplotlib.figure.Figure at 0x7f1e51fdb3d0>






    





<matplotlib.figure.Figure at 0x7f1e51fdb1d0>



In [2]:

    
def make_curves(model, *args):
    curves = None
    for c in args:
        channel = m.monitor.channels[c]
        c = c[0].upper() + c[1:]
        if not curves:
            curves = hl.Curve(zip(channel.epoch_record, channel.val_record),group=c)
        else:
            curves += hl.Curve(zip(channel.epoch_record, channel.val_record),group=c)
    return curves



In [3]:

    
m = pylearn2.utils.serial.load(
    "/disk/scratch/neuroglycerin/models/opencv_integrated_v3_reg_ranged_recent.pkl")



In [4]:

    
make_curves(m,"valid_objective","valid_y_nll","train_y_nll")









    Out[4]:

The train_y_nll, valid_y_nll and valid_objective show massive overfitting.



In [26]:

    
m.layer_names









    Out[26]:





{'h5', 'h6', 'parallel', 'y'}



In [70]:

    
channel = m.monitor.channels["valid_y_nll"]
hl.Curve(zip(channel.epoch_record, channel.val_record),label="valid_y_nll")









    Out[70]:



In [71]:

    
channel = m.monitor.channels["valid_y_nll"]
plt.plot(channel.epoch_record, channel.val_record)









    Out[71]:





[<matplotlib.lines.Line2D at 0x7f3ec1c3ff50>]

Hard to see whether it is still learning...



In [72]:

    
ch1 = m.monitor.channels["valid_y_nll"]
ch2 = m.monitor.channels["train_y_nll"]
hl.Curve(zip(ch1.epoch_record[-40:], ch1.val_record[-40:]),label="valid_y_nll")









    Out[72]:



In [73]:

    
hl.Curve(zip(ch2.epoch_record[-40:], ch2.val_record[-40:]),label="train_y_nll")









    Out[73]:



In [74]:

    
m = pylearn2.utils.serial.load(
    "/disk/scratch/neuroglycerin/models/continue_hopeful_recent.pkl")



In [75]:

    
make_curves(m,"valid_objective","valid_y_nll","train_y_nll")









    Out[75]:



In [ ]: