In [28]:

    

import numpy
import pandas
import json


    

In [32]:

    

# Load the data from file, make it into a pretty pandas Dataframe
data = pandas.read_json('samples/target.json').T
data['name'] = data.index
data.reset_index(inplace=True, drop=True) # reset index to numbers, drop extra index column
data.head()


    

    
Out[32]:






  

    

      

      
class
      
name
    
  
  

    

      
0
      
NIGHT
      
snapshot-2017-07-22_20-20-01.jpg
    
    

      
1
      
NIGHT
      
snapshot-2017-07-22_20-30-01.jpg
    
    

      
2
      
NIGHT
      
snapshot-2017-07-22_20-40-01.jpg
    
    

      
3
      
NIGHT
      
snapshot-2017-07-22_20-50-01.jpg
    
    

      
4
      
NIGHT
      
snapshot-2017-07-22_21-00-01.jpg
    
  

In [33]:

    

# Getting an idea of how the data is distributed
data['class'].value_counts()


    

    
Out[33]:





NIGHT    435
WET      280
DRY       67
Name: class, dtype: int64

In [34]:

    

from roofcam.classifier import classify_wet_or_dry
data['prediction'] = data.apply(lambda row: classify_wet_or_dry("samples/" + row['name']), axis=1)
data.head()


    

    
Out[34]:






  

    

      

      
class
      
name
      
prediction
    
  
  

    

      
0
      
NIGHT
      
snapshot-2017-07-22_20-20-01.jpg
      
NIGHT
    
    

      
1
      
NIGHT
      
snapshot-2017-07-22_20-30-01.jpg
      
NIGHT
    
    

      
2
      
NIGHT
      
snapshot-2017-07-22_20-40-01.jpg
      
NIGHT
    
    

      
3
      
NIGHT
      
snapshot-2017-07-22_20-50-01.jpg
      
WET
    
    

      
4
      
NIGHT
      
snapshot-2017-07-22_21-00-01.jpg
      
WET
    
  

In [35]:

    

data['prediction'].value_counts()


    

    
Out[35]:





WET      405
NIGHT    341
DRY       26
ERROR     10
Name: prediction, dtype: int64

In [84]:

    

# Drop error rows, we're not that interested in them (errors are caused by faulty images)
data = data[data['prediction'] != "ERROR"]

# Drop NIGHT rows
# data = data[data['prediction'] != "NIGHT"]
# data = data[data['class'] != "NIGHT"]


    

In [85]:

    

correct = len(data[data['prediction'] == data['class']])
correct_percent = float(correct)/len(data) * 100
print "TOTAL", len(data)
print "CORRECT", correct, "({0:0.0f}%)".format(correct_percent)
print "INCORRECT", len(data) - correct, "({0:0.0f}%)".format(100 - correct_percent)


    

    




TOTAL 315
CORRECT 282 (90%)
INCORRECT 33 (10%)

In [86]:

    

from pandas_ml import ConfusionMatrix

y_true=data['class']
y_pred=data['prediction']
confusion_matrix = ConfusionMatrix(y_true, y_pred)
print("Confusion matrix:\n%s" % confusion_matrix)


    

    




Confusion matrix:
Predicted  DRY  WET  __all__
Actual                      
DRY         26   33       59
WET          0  256      256
__all__     26  289      315

In [87]:

    

%matplotlib inline
import matplotlib.pyplot as plt

confusion_matrix.plot()


    

    
Out[87]:





<matplotlib.axes._subplots.AxesSubplot at 0x119ed2410>






    

In [ ]: