def Assign(i,lhs,rhs):
## Preprocessing:
ll = base.load('assign.ll')
## Processing:
#-------------------------------------------------------------------
code = ''.join(ll)
code = code.replace('<i>','%s'%i)
code = code.replace('${memlhs}',lhs)
code = code.replace('${memrhs}',rhs)
return code
def If(i,cond,if_body,else_body):
## Preprocessing:
ll = base.load('If.ll')
## Processing:
#-------------------------------------------------------------------
code = ''.join(ll)
code = code.replace('<i>','%s'%i)
code = code.replace('${cond}',cond)
code = code.replace('${if_body}',if_body)
code = code.replace('${else_body}',else_body)
return code
def Num(i, Num, begin_annotation=";START"):
## Preprocessing:
ll = base.load("Num.ll", begin_annotation)
## Processing:
# -------------------------------------------------------------------
code = "".join(ll)
code = code.replace("<i>", "%s" % i)
code = code.replace("${Num}", str(Num))
out = "memz_%s" % i
return out, code
def corr(self):
self.set_subset('silhouette')
df = self.acc_single()
df['dataset'] = 'silhouette'
human = base.load(pref='preds', exp=self.exp, suffix='human')
df['human_accuracy'] = np.nan
n = len(df.dataset.unique())
mean_acc = human[human.kind == 'silhouette'].groupby('no').acc.mean()
df.loc[:, 'human_accuracy'] = mean_acc.tolist() * n
df = df[df.sel]
df.model_accuracy = df.model_accuracy.astype(int)
sns.set_palette(sns.color_palette('Set2')[1:])
self._corr(df, 'all')
def While(i,cond,test_body,body):
## Preprocessing:
ll = base.load('while.ll')
## Processing:
#-------------------------------------------------------------------
code = ''.join(ll)
code = code.replace('<i>','%s'%i)
code = code.replace('${cond}',cond)
code = code.replace('${test_body}',test_body)
code = code.replace('${body}',body)
return code
def Module(i,modName,body,begin_annotation = ';START'):
## Preprocessing:
ll = base.load('module.ll',begin_annotation)
## Processing:
#-------------------------------------------------------------------
code = ''.join(ll)
code = code.replace('<i>','%s'%i)
code = code.replace('${modname}',modName)
code = code.replace('${body}',body)
return code
def corr(self):
self.set_subset('silhouette')
df = self.acc_single()
df['dataset'] = 'silhouette'
human = base.load(pref='preds', exp=self.exp, suffix='human')
df['human_accuracy'] = np.nan
n = len(df.dataset.unique())
mean_acc = human[human.kind=='silhouette'].groupby('no').acc.mean()
df.loc[:,'human_accuracy'] = mean_acc.tolist() * n
df = df[df.sel]
df.model_accuracy = df.model_accuracy.astype(int)
sns.set_palette(sns.color_palette('Set2')[1:])
self._corr(df, 'all')
def logicNot(i,x,begin_annotation = ';START'):
## Preprocessing:
ll = base.load('logicNot.ll',begin_annotation)
## Processing:
#-------------------------------------------------------------------
code = ''.join(ll)
code = code.replace('<i>','%s'%i)
code = code.replace('${memx}',x)
out = 'memz_%s'%i
return out,code
def pred_corr(self, value='accuracy', method='corr'):
human = base.load(pref='preds', exp=self.exp, suffix='human')
human = human.groupby(['kind', 'no']).acc.mean()
dfs = []
for subset in ORDER:
self.set_subset(subset)
df = self._pred_corr(human.loc[subset], value=value, method=method)
df['dataset'] = subset
dfs.append(df)
df = pandas.concat(dfs, ignore_index=True)
print(df.groupby('dataset').mean())
if self.task == 'run':
self.plot_single(df, 'pred_corr')
return df
def logical(i,x,y,op,begin_annotation = ';START'):
## Preprocessing:
ll = base.load('logical.ll',begin_annotation)
## Processing:
#-------------------------------------------------------------------
code = ''.join(ll)
code = code.replace('<i>','%s'%i)
code = code.replace('${memx}',x)
code = code.replace('${memy}',y)
code = code.replace('${op}', opMap[op])
out = 'memz_%s'%i
return out,code
def pred_corr(self, value='accuracy', method='corr'):
human = base.load(pref='preds', exp=self.exp, suffix='human')
human = human.groupby(['kind', 'no']).acc.mean()
dfs = []
for subset in ORDER:
self.set_subset(subset)
df = self._pred_corr(human.loc[subset], value=value, method=method)
df['dataset'] = subset
dfs.append(df)
df = pandas.concat(dfs, ignore_index=True)
print(df.groupby('dataset').mean())
if self.task == 'run':
self.plot_single(df, 'pred_corr')
return df
def compare(i,x,y,op,begin_annotation = ';START'):
## Preprocessing:
ll = base.load('comp.ll',begin_annotation)
## Processing:
#-------------------------------------------------------------------
code = ''.join(ll)
code = code.replace('<i>','%s'%i)
code = code.replace('${memx}',x)
code = code.replace('${memy}',y)
code = code.replace('${opi}', opMap[op][0])
code = code.replace('${opf}', opMap[op][1])
out = 'memz_%s'%i
return out,code
def behav(self):
self.model_name = 'behav'
human = base.load(pref='preds', exp=self.exp, suffix='human')
sil = pandas.read_csv('snodgrass/data/sil_human_acc.csv', header=None) / 100.
sil2 = human.groupby(['kind', 'no']).acc.mean()
corr = np.corrcoef(sil.values.ravel(), sil2['silhouette'].values)[0,1]
cons = 1 - scipy.spatial.distance.sqeuclidean(sil.values.ravel(),
sil2['silhouette'].values) / 260
sel, _ = self.filter_synset_ids()
human = human[human.synset_id.isin(sel)]
if self.task == 'run':
sns.factorplot('kind', 'acc', data=human, units='subjid',
kind='bar', color=self.colors['shape'])
self.show(pref='acc')
self.html.writetable(human.groupby('kind').acc.mean())
self.html.write('<p>Correlation old-new: {:.2f}</p>'.format(corr))
self.html.write('<p>Consistency old-new: {:.2f}</p>'.format(cons))
return human
def cprint(i,args):
## Preprocessing:
ll = base.load('print.ll')
## Processing:
#-------------------------------------------------------------------
code = ''.join(ll)
arglist = ''
loader = ''
for j,arg in enumerate(args):
loader += '%%arg%s_%s.TYPE = load i32* %%%s.TYPE'%(j+1,i,arg)
loader += '%%arg%s_%s.n = load i64* %%%s.n'%(j+1,i,arg)
arglist+=',i32 %%arg%s_%s.TYPE'%(j+1,i)
arglist+=',i64 %%arg%s_%s.n'%(j+1,i)
print len(args)
code = loader + code.replace('${args}','i32 %s %s'%(len(args),arglist))
return code
def behav(self):
self.model_name = 'behav'
human = base.load(pref='preds', exp=self.exp, suffix='human')
sil = pandas.read_csv('snodgrass/data/sil_human_acc.csv',
header=None) / 100.
sil2 = human.groupby(['kind', 'no']).acc.mean()
corr = np.corrcoef(sil.values.ravel(), sil2['silhouette'].values)[0, 1]
cons = 1 - scipy.spatial.distance.sqeuclidean(
sil.values.ravel(), sil2['silhouette'].values) / 260
sel, _ = self.filter_synset_ids()
human = human[human.synset_id.isin(sel)]
if self.task == 'run':
sns.factorplot('kind',
'acc',
data=human,
units='subjid',
kind='bar',
color=self.colors['shape'])
self.show(pref='acc')
self.html.writetable(human.groupby('kind').acc.mean())
self.html.write('<p>Correlation old-new: {:.2f}</p>'.format(corr))
self.html.write('<p>Consistency old-new: {:.2f}</p>'.format(cons))
return human
import sys
import base
for k, v in base.load().iteritems():
setattr(sys.modules[__name__], k, v)
import cv2
from segmentation import kmeans
from models import Image
from segmentation import remove_red_cells
from segmentation import background_removal
from base import load
from segmentation import otsu
from matplotlib import pyplot as plt
import numpy as np
from math import sqrt
path = 'bases/ALL_IDB1/'
images = load(path)
#images = [images[3]]
for image in images:
lab_image = cv2.cvtColor(image.image, cv2.COLOR_BGR2LAB)
hsv_image = cv2.cvtColor(image.image, cv2.COLOR_BGR2HSV)
yuv_image = cv2.cvtColor(image.image, cv2.COLOR_BGR2YUV)
gray_image = cv2.cvtColor(image.image, cv2.COLOR_BGR2GRAY)
blue, green, red = cv2.split(image.image)
hue, saturation, value = cv2.split(hsv_image)
l, a, b = cv2.split(lab_image)
y, u, v = cv2.split(yuv_image)
'''
#cv2.imwrite('temp/temp/blue_' + image.name, blue)
cv2.imwrite('temp/temp/green_' + image.name, green)
#cv2.imwrite('temp/temp/red_' + image.name, red)
cv2.imwrite('temp/temp/hue_' + image.name, hue)
cv2.imwrite('temp/temp/saturation_' + image.name, saturation)
#cv2.imwrite('temp/temp/value_' + image.name, value)
def _makeobj(self, key):
obj = self._model()
r = redis_connection()
res = r.get(key)
if res: obj.__dict__.update( load(res))
return obj
import sys
import base
for k, v in base.load().iteritems():
setattr(sys.modules[__name__], k, v)