Egor Beliaev hcl14

## peopledetect.cpp
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/objdetect/objdetect.hpp"
#include "opencv2/highgui/highgui.hpp"

#include <stdio.h>
#include <string.h>
#include <ctype.h>
//#include "openCVUtils.h"

using namespace cv;

## newton_tensorflow.py
# Newton's method in Tensorflow

# 'Vanilla' N.m. intended to work when loss function to be optimized is convex.
# One-layer linear network without activation is convex.
# If activation function is monotonic, the error surface associated with a single-layer model is convex.

# In other cases, Hessian will have negative eigenvalues in saddle points and other non-convex places of the surface
# To fix that, you can try different methods. One of those approaches is to do eigendecomposition of H and invert negative eigenvalues,
# making H "pushing out" in those directions, as described in this paper: Identifying and attacking the saddle point problem in high-dimensional non-convex optimization (https://papers.nips.cc/paper/5486-identifying-and-attacking-the-saddle-point-problem-in-high-dimensional-non-convex-optimization.pdf)

## newton_tensorflow_iris.py
# Newton's method in Tensorflow

# 'Vanilla' N.m. intended to work when loss function to be optimized is convex.
# One-layer linear network without activation is convex.
# If activation function is monotonic, the error surface associated with a single-layer model is convex.

# In other cases, Hessian will have negative eigenvalues in saddle points and other non-convex places of the surface
# To fix that, you can try different methods. One of those approaches is to do eigendecomposition of H and invert negative eigenvalues,
# making H "pushing out" in those directions, as described in this paper: Identifying and attacking the saddle point problem in high-dimensional non-convex optimization (https://papers.nips.cc/paper/5486-identifying-and-attacking-the-saddle-point-problem-in-high-dimensional-non-convex-optimization.pdf)

## newton_tensorflow_sklearn_digits.py
# Newton's method in Tensorflow

# WARNING! This code is memory and computationally intensive, better run it on GPU
# having bigger dimensionality increases computing time significantly
# Original dataset is passable on GTX 1050 GPU, but if you have time/memory problems, uncomment PCA compression
# Also, you can probably remove line 159 (hessian fixing) if you use PCA

# 'Vanilla' N.m. intended to work when loss function to be optimized is convex.
# One-layer linear network without activation is convex.
# If activation function is monotonic, the error surface associated with a single-layer model is convex.

## tf_second_order_incremental.py
# Layer-wise training neural network with second order (Newton).
# New layer is added on each iteration and optimized with Newton Method.
# And example of Tensorflow eager execution
# Combines gradiets, hessian, and call to Optimizer

# Might contain logical errors though, so think yourself when adapting this code


# Newton's method in Tensorflow

## database.txt
Which color is normally a cat?;Black
How tall was the longest man on earth?;272 cm
Is the earth round?;Yes
Which color is normally a cat?;Black
How tall was the longest man on earth?;272 cm
Is the earth round?;Yes
Which color is normally a cat?;Black
How tall was the longest man on earth?;272 cm
Is the earth round?;Yes
Which color is normally a cat?;Black

## qlearn_simple.py

# Simple example of Q-learning inability to go in loops
# Though it is strictly forbibben by the code (line 101),
# but you can comment out that logic and see that algorithm just becomes less stable
# The reason is that loop is impossible in this setup,
# as only a single Q-value exists for each position on the map


import numpy as np
np.random.seed(0)

## flask_app.py
# create flask app to be run by tornado process

# process-specific globals
import global_vars

from flask import Flask, request, Response, json, abort, jsonify

# oridnary (non-flask) json if needed
import json as json2


## 1_tkinter_separate.py
# My answer here: https://stackoverflow.com/questions/52793096/reload-and-zoom-image/52818151#52818151

from PIL import ImageTk, Image

from scipy.ndimage import rotate
from scipy.misc import imresize
import numpy as np
import time


## app.py
## Author: Victor Dibia
## Load hand tracking model, spin up web socket and web application.


from utils import detector_utils as detector_utils
from utils import object_id_utils as id_utils
import cv2
import tensorflow as tf
import multiprocessing
from multiprocessing import Queue, Pool
	#include "opencv2/imgproc/imgproc.hpp"
	#include "opencv2/objdetect/objdetect.hpp"
	#include "opencv2/highgui/highgui.hpp"

	#include <stdio.h>
	#include <string.h>
	#include <ctype.h>
	//#include "openCVUtils.h"

	using namespace cv;
	# Newton's method in Tensorflow

	# 'Vanilla' N.m. intended to work when loss function to be optimized is convex.
	# One-layer linear network without activation is convex.
	# If activation function is monotonic, the error surface associated with a single-layer model is convex.

	# In other cases, Hessian will have negative eigenvalues in saddle points and other non-convex places of the surface
	# To fix that, you can try different methods. One of those approaches is to do eigendecomposition of H and invert negative eigenvalues,
	# making H "pushing out" in those directions, as described in this paper: Identifying and attacking the saddle point problem in high-dimensional non-convex optimization (https://papers.nips.cc/paper/5486-identifying-and-attacking-the-saddle-point-problem-in-high-dimensional-non-convex-optimization.pdf)
	# Newton's method in Tensorflow

	# WARNING! This code is memory and computationally intensive, better run it on GPU
	# having bigger dimensionality increases computing time significantly
	# Original dataset is passable on GTX 1050 GPU, but if you have time/memory problems, uncomment PCA compression
	# Also, you can probably remove line 159 (hessian fixing) if you use PCA

	# 'Vanilla' N.m. intended to work when loss function to be optimized is convex.
	# One-layer linear network without activation is convex.
	# If activation function is monotonic, the error surface associated with a single-layer model is convex.
	# Layer-wise training neural network with second order (Newton).
	# New layer is added on each iteration and optimized with Newton Method.
	# And example of Tensorflow eager execution
	# Combines gradiets, hessian, and call to Optimizer

	# Might contain logical errors though, so think yourself when adapting this code



	# Newton's method in Tensorflow
	Which color is normally a cat?;Black
	How tall was the longest man on earth?;272 cm
	Is the earth round?;Yes
	Which color is normally a cat?;Black
	How tall was the longest man on earth?;272 cm
	Is the earth round?;Yes
	Which color is normally a cat?;Black
	How tall was the longest man on earth?;272 cm
	Is the earth round?;Yes
	Which color is normally a cat?;Black

	# Simple example of Q-learning inability to go in loops
	# Though it is strictly forbibben by the code (line 101),
	# but you can comment out that logic and see that algorithm just becomes less stable
	# The reason is that loop is impossible in this setup,
	# as only a single Q-value exists for each position on the map


	import numpy as np
	np.random.seed(0)
	# create flask app to be run by tornado process

	# process-specific globals
	import global_vars

	from flask import Flask, request, Response, json, abort, jsonify

	# oridnary (non-flask) json if needed
	import json as json2
	# My answer here: https://stackoverflow.com/questions/52793096/reload-and-zoom-image/52818151#52818151

	from PIL import ImageTk, Image

	from scipy.ndimage import rotate
	from scipy.misc import imresize
	import numpy as np
	import time
	## Author: Victor Dibia
	## Load hand tracking model, spin up web socket and web application.


	from utils import detector_utils as detector_utils
	from utils import object_id_utils as id_utils
	import cv2
	import tensorflow as tf
	import multiprocessing
	from multiprocessing import Queue, Pool