images = glob.glob('../bee_project/*.jpg')
for f_name in images:
if count >= batch_size:
X_train = numpy.array(data,dtype='float32')
y_train = numpy.array(labels,dtype='int32')
data = []
labels = []
count = 0
yield (X_train,y_train)
im = Image.open(f_name,mode='r')
dat = numpy.asarray(im).astype('float32') / 255
im.close()
dat = numpy.rollaxis(dat,2)
data.append(dat)
#hacky way of getting classification from filename
jpg_idx = f_name.find('.jpg')
labels.append(int(f_name[jpg_idx-1:jpg_idx]))
count+=1
yield (X_train,y_train)
for X_epoch,y_epoch in generate_data():
#fit the batchsize to our model
convNet.partial_fit(X_epoch,y_epoch)
with open('convNet.pickle','wb') as f:
#save our model to a file
pickle.dump(convNet,f,-1)
