prerakmody

"""
References

1. Calculating area under the curve (AUC) using Riemann Sum - https://en.wikipedia.org/wiki/Riemann_sum
2. https://scikit-learn.org/stable/modules/generated/sklearn.metrics.auc.html#sklearn.metrics.auc
 - it uses np.trapz()
    - Code: https://github.com/numpy/numpy/blob/v1.22.0/numpy/lib/function_base.py#L4686-L4797
3. Other methods implemented in TFlow
    - https://wikidocs.net/24605
    - Note: This method only works right when we provide a list x that has evenly spaced values

prerakmody

"""
OG Ref: https://github.com/tensorflow/probability/issues/620

Goals
 - To experiment with DNNs built using Flipout layers in both eager and non-eager mode
    - Eager mode allows for debugging using tf.print()
    - Non-eager mode is supposed to be faster and less memory consuming since its pre-computes functions in a graph 
 - Models will be made using functional API and the dataset will not use tf.data.Dataset

Notes

prerakmody

"""
Understanding Entropy and Mutual information (epistemic uncertainty) with Stochastic Forward Passes (=M)
Binary Classification
 - y = {(p1, (1-p1)), (pn, (1-pn)), ... (pn, (1-pn))}
 - Ent  = -(pbar.log(pbar) + (1-pbar).log(1-pbar))
 - MIF  = Ent + avg_M(p1.log(p1) + ... + pn.log(pn) + (1-p1).log(p1) + ... + pn.log(1-pn))

 - Case1 
 -- y = {(1,0), (1,0), ..., (1,0)}
 -- pbar = (1,0)

prerakmody

import numpy as np
import matplotlib
import matplotlib.pyplot as plt

cmap = matplotlib.cm.get_cmap('magma')
f,axarr = plt.subplots(1,3)

x   = np.arange(0,1,0.01)
img = np.zeros((len(x),len(x),4))
for x_id, x_ in enumerate(x):img[x_id,:,:] = np.repeat(np.expand_dims(np.array(cmap(x_)), axis=0), len(x), axis=0)

prerakmody

import skimage
import skimage.measure
import plotly.graph_objects as go

voxel_mask = #
verts, faces, _, _ = skimage.measure.marching_cubes(voxel_mask, step_size=1)

fig = go.Figure(data=[go.Mesh3d(x=verts[:,0], y=verts[:,1], z=verts[:,2], i=faces[:,0],j=faces[:,1],k=faces[:,2])])
fig.write_html('tmp.html') # in VS-Code install the "HTML Preview" extension (Analytic Signal Limited)
fig.show()

prerakmody

"""
pip install tensorflow
pip install tf2onnx keras2onnx onnxmltools
"""

import os
import pdb
import json
import traceback
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3" 

prerakmody

"""
This method uses the max pooling operation to get the boundary around a binary mask
Can be used in segmentation tasks where one wants to specific a loss for the boundary pixels/voxels 
"""

import os
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3" 
import nrrd
import numpy as np
import tensorflow as tf

prerakmody

import os
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3" 

import tensorflow as tf # v2.4 with CUDA11.0.0 & CuDNN8.0.0
import tensorflow_probability as tfp # v0.12.1

if len(tf.config.list_physical_devices('GPU')):tf.config.experimental.set_memory_growth(tf.config.list_physical_devices('GPU')[0], True)

x = tf.random.normal((1,100,100,100,10)) # B, H,W,D,C
layer = tfp.layers.Convolution3DFlipout(filters=16, kernel_size=1) 

prerakmody

"""
Expected Calibration Error for semantic segmentation tasks
"""
import pdb
import nrrd  # pip install pynrrd
import traceback
import numpy as np
from pathlib import Path
import matplotlib.pyplot as plt

prerakmody

import numpy as np
import matplotlib.pyplot as plt

f,axarr = plt.subplots(1,2)

# Step 1
x       = [0.01, 0.02, 0.03, 0.04] + np.arange(0.05,1.05,0.05).tolist()
y_ce    = [-np.log(each) for each in x]
y_focal1 = [-1*(1 - each)**1*np.log(each) for each in x]
y_focal2 = [-1*(1 - each)**2*np.log(each) for each in x]