An Overview of Keras Deep Learning

We may definitely know machine taking in, a branch in software engineering that reviews the plan of calculations that can learn. Today, you will center around profound taking in, a subfield of machine discovering that is an arrangement of calculations that is enlivened by the structure and capacity of the mind. These calculations are typically called Artificial Neural Networks (ANN). Profound learning is one of the most sizzling fields in information science with numerous contextual investigations with great outcomes in mechanical autonomy, picture acknowledgment and Artificial Intelligence (AI). 

A standout amongst the most intense and simple to-utilize Python libraries for creating and assessing profound learning models is Keras It wraps the proficient numerical calculation libraries Theano and TensorFlow. The benefit of this is for the most part that you can begin with neural systems in a simple and fun way.

Linear Regression 

A regression issue implies we need to foresee a genuine esteemed yield. For instance, how about we utilize the (normal) house valuing illustration where we just have 1 trademark (include), the territory of the house, and we need to foresee the house cost. Given an arrangement of area(X)/house-price(Y) sets (dataset) we can plot the accompanying diagram (where every orange speck is a region/house-value combine): 

From essential straight variable based math, we realize that the condition of the line is in the shape Y = θ0 + θ1*X. This is known as the theory and what we are searching for are the qualities for θ0 and θ1 (called the parameters). On the off chance that we locate a decent arrangement of θ0, θ1 then we will have the capacity to anticipate a house value given the region of any house. To do that, first we have to characterize the mistake/cost which is the distinction of our speculation from the genuine esteem: 

There are numerous approaches to figure the cost work with a standout amongst the most famous to be the mean squared mistake. At that point we should simply to begin from a (the greater part of the circumstances irregular) beginning task of θ0, θ1 and utilizing an iterative enhancement calculation (like Gradient Descent) to limit the cost of our speculation by emphasizing over our dataset. 

import pandas as pds

dataframeX = pds.read_csv('No-show-Issue-Comma-300k.csv', usecols=[0, 1, 4, 6, 7, 8, 9, 10, 11, 12, 13])
dataframeY = pds.read_csv('No-show-Issue-Comma-300k.csv', usecols=[5])

print(dataframeX.head())
print(dataframeY.head())

Calculated Regression 

Be that as it may, imagine a scenario in which we don't need a genuine esteemed yield and we might want a likelihood. We can utilize Logistic Regression where the speculation yields genuine esteemed numbers from 0 to 1. To do that we can utilize the Sigmoid capacity to outline genuine esteemed number of our speculation to the (0, 1) interim: 

Sigmoid capacity 

We would  be able to adjust the cost capacity and utilize a similar enhancement calculations to discover a speculation that limits the cost. 

Neural Networks 

Be that as it may, the greater part of this present reality issues are not straightly distinguishable. For reasons unknown interfacing little units that do calculated relapse between them is a standout amongst the most computationally productive approaches to figure non-straight speculation: 

Each hub in the diagram, aside from input layer, speaks to straightforward strategic relapse with inputs the approaching edges and yields the outcoming edges. The furthest right hub, the yield layer, will give us a last likelihood given the info qualities (highlights). 

Basic issues and approval set 

The two most basic issues are underfitting and overfitting (with the last to be more famous). Underfitting happens when your theory does not fit the information all around ok, overfitting happens when the speculation is too firmly fit the preparation dataset and does not sum up well to new concealed information. To assess our model, we split our dataset into the preparation and approval set. At that point we utilize the preparation set to fit our model and test it with the inconspicuous approval set.

# 1
import numpy as np
seed = 7
np.random.seed(seed)

# 2
from keras.models import Sequential
from keras.layers import Dense
model = Sequential()
model.add(Dense(12, input_shape=(11,), init='uniform', activation='sigmoid'))
model.add(Dense(12, init='uniform', activation='sigmoid'))
model.add(Dense(12, init='uniform', activation='sigmoid'))
model.add(Dense(1, init='uniform', activation='sigmoid'))
model.summary()

# 3
import keras
tbCallBack = keras.callbacks.TensorBoard(log_dir='/tmp/keras_logs', write_graph=True)

# 4
model.compile(loss='mean_squared_error', optimizer='adam', metrics=['accuracy'])
model.fit(dataframeX.values, dataframeY.values, epochs=9, batch_size=50,  verbose=1, validation_split=0.3, callbacks=[tbCallBack])