tensorflowmodel.py

# Execute this cell to be sure to have a compatible TF (2.0) and TFP (0.8) version. 
# If you are bold you can skip this cell. 
try: #If running in colab 
  import google.colab
  !pip install tensorflow==2.0.0
  !pip install tensorflow_probability==0.8.0
except:
  print('Not running in colab')

try: #If running in colab 
    import google.colab
    IN_COLAB = True 
    %tensorflow_version 2.x
except:
    IN_COLAB = False
    
    
#Imports
import tensorflow as tf
if (not tf.__version__.startswith('2')): #Checking if tf 2.0 is installed
    print('Please install tensorflow 2.0 to run this notebook')
print('Tensorflow version: ',tf.__version__, ' running in colab?: ', IN_COLAB)

import matplotlib.pyplot as plt
import numpy as np
from sklearn.model_selection import train_test_split
import tensorflow_probability as tfp

%matplotlib inline
plt.style.use('default')

tfd = tfp.distributions
tfb = tfp.bijectors
print("TFP Version", tfp.__version__)
print("TF  Version",tf.__version__)
np.random.seed(42)
tf.random.set_seed(42)

#Working with a TFP normal distribution
# 10          20       30        40       50         55
#123456789012345678901234567890123456789012345678901234
import tensorflow_probability as tfp
tfd = tfp.distributions
d = tfd.Normal(loc=3, scale=1.5)           #A
x = d.sample(2) # Draw two random points.    #B
px = d.prob(x) # Compute density/mass.       #C
print(x)
print(px)
#A create a 1D Normal distribution with mean 3 and standard deviation 1.5
#B sample 2 realizations from the Normal distribution
#C compute the likelihood for each of the two sampled values under the defined Normal distribution

dist = tfd.Normal(loc=1.0, scale=0.1)
print('sample   :', dist.sample(3).numpy()) #Samples 3 numbers
print('prob     :',dist.prob((0,1,2)).numpy()) #Calculates the probabilities for positions 0,1,2
print('log_prob :',dist.log_prob((0,1,2)).numpy()) #Same as above just log
print('cdf      :',dist.cdf((0,1,2)).numpy()) #Calculates the cummulative distributions
print('mean     :',dist.mean().numpy()) #Returns the mean of the distribution
print('stddev   :',dist.stddev().numpy())