eChapters

In this chapter of machine learning algorithms we will cover the basic algorithms of machine learning with their implementation using dataset and discuss their pro’s and cons of algorithms for better understanding and how to use them in any project or with any dataset to get an correct result which can useful for research or study.

Key Component

We defined a data set composed of audio, images and binary labels that provided an understanding of how we could train a model from snippets to classifications. This sort of problem is one of many kinds of machine learning problems when we try and predict a specified unknown label given known inputs, provided that a dataset of examples for which labels are known is called supervised learning. Let’s discus what are the major key components we have to get there

In this chapter we will cover all the basics libraries and some frameworks of machine learning, deep learning and computer vision. We will discuss all about their use with an example and instruction which will states the commands to install to those libraries for building any project in our system. So let’s begin

What is a Library?

Library is an aggregate of features and methods that allow you to take many actions without getting your own code written.

For instance, the libraries that you have to know are if you work with data, numpy, scipy, pandas, etc. They have very simple data processing functions that save you time for small tricks.

In this chapter we will talk about the tools which are required to make project in artificial intelligence. All this bellow discussed tools are uses python programming language for making any project in any respective field. We have discussed about how anyone can install and use the tools to built a real time project in artificial Intelligence.

Using Python IDE

To download and install Python visit the official website of Python https://www.python.org/downloads/and choose your version. We have chosen Python version as per requirement all the codes in this book were on python version 3.6 as you can see in the Fig. 7.1.

In this chapter we will look at the Expert system which can guide and assist an individual once it is provided with the enough Knowledge to do so. Here we will discuss about expert system and its architecture with proper steps of making an expert system in any field by several tools which are available to do so. For expert systems the most important is Knowledge Engineering which is the collection of Knowledge from an expert and Engineering it in such a form that a machine could understand it.

Introduction

One of the pioneering fields of AI study is Expert Systems (ES). It is proposed by researchers in the computer science department of Stanford University. An expert system is a computer system which emulates a human expert’s decision-making skills. Expert systems are developed by reasoning through information structures, representing laws, rather than through traditional procedural code, in order to overcome complex problems.

Expert Programs use professional expertise systematically to address human expert problems. An expert is an individual with experience in a specific field. The expert has expertise or special credentials which most people do not learn or care about. An expert can solve or solve problems more effectively than people can solve. Expert System actually emulates all these qualities of expert and solve the problem in a better way and fast.

Today, Expert System (ES) is used in the fields of finance, research, engineering, technology, medicine and many others where there is a well-defined problem area. The fundamental principle is that an expert method should also define the steps to illustrate why a problem can be solved. Unless an individual can justify his reasoning for Expert System.

Introduction

Agriculture has played an important role in the global economy in recent years. The further growth of the population is leading to a gradual decrease in the area of the agricultural land and further growing stresses on the farm system. Agricultural food production methods are rising in demand for successful and healthy. Innovative sensoring and motive technologies and advanced information and communications systems and Artificial Intelligence must supplement conventional agricultural management methods to speed up farm productivity growth more accurately, thereby promoting the production of high-quality and high-performance agriculture. Computer vision monitoring systems have become valuable tools in farm operations over the past decades, and their use has increased significantly.

What is Computer Vision?

Computer vision can be characterized as a field of AI for extracting information from digital images. The type of data obtained from an image may be different from recognition, navigational space measurements or applications of increased reality.

The applications of computer vision can also be described. Computer vision builds algorithms that recognize and use the quality of pictures for other purposes.

Deep learning emerged as a serious contender in the field from a 10-year explosive computational development. Deep learning is therefore a special form of machine learning, the algorithms of which are based on the structure and function of the human brain.

Machine learning VS Deep learning

Deep learning is arguably the most efficient form of machine learning. This is so important because they know the best way to solve the problem and how to fix everything. The following is a description of deep learning and machine Learning:

As we know there are different branches of artificial Intelligence and machine Learning of one of them which has become a most advance and widely used branch of artificial Intelligence in every field of this generation whether it may be either electronics, civil, mechanics, Bioengineering, medical or agriculture etc., In this chapter we will study about machine learning and its types with the most common algorithms of machine learning which are being mostly used in different scenarios and their implementation using python language has been discussed in the implementation part of this book. Here we will also see where machine learning can be used.

i. Introduction

Machine learning is a subset of Artificial Intelligence and one of the most important part of artificial Intelligence. It is a field of Computer science which is totally differs from traditional computational approaches. In traditional approaches, a set of instruction is explicitly provided to the algorithm to solve any problem. Whereas in Machine Learning is a domain of an AI which uses a statistical analysis and can enables a system to always learn from its data without any external inputs in order to give an essential output. Due to this machine learning makes computers in advance to make model which provide decision based on input data.

Machine Learning involves a lot of algorithms which usually based on their requirements and functionalities to use where and when. To make a machine learning model we always require the data on which model can be made. Machine learning uses the algorithm which continuously learns from the data available as training data. A machine learning model is generated after the training of algorithm from the data. After the training completes, when trained model is provided with other real-life data then you can get an essential outcome. For example, in figure 3.1 Machine learning Model we use a predictive algorithm and make a model by providing it with some data, and later on we can receive prediction’s based on real data as input.

As we have seen how artificial Intelligence can be useful in agriculture but the nest question always come in mind is that from where and how we can start to learn and what are the model of learning an artificial Intelligence. As Artificial Intelligence is a huge part of Computer Science which means we need an language as medium to perform it and we always prefer Python for that which is very dynamic and easy to learn. In this chapter we will learn about python basics so that in further implementation of artificial Intelligence it will be helpful to us.

I. Introduction

Python is a remarkable language that can claim to be easy and strong. You will be amazed to see how simple it is not to focus on the language you are programming, but on solving the problem. Guido van Rossum was the creator of the Python language which is named after the BBC show “Monty Python’s flying Circus”.

Python is formally introduced as:

“Python is an easy to learn, powerful programming language. It has efficient high-level data structures and a simple but effective approach to object-oriented programming. Python’s elegant syntax and dynamic typing, together with its interpreted nature, make it an ideal language for scripting and rapid application development in many areas on most platforms.”

The artificial intelligence (AI) world is quickly growing as it reaches several different industries. You currently will see AI used for many different purposes from agriculture to automotive; with the passage of time you will see even more of it. Agriculture is one of the most important branches of the IT industry. Agriculture is an important industry and is a big part of our economy’s base. In terms of annual sales, the agricultural industry contributes to our economy almost 17%.AI is becoming a technical advancement that boost and protects crop yield, as climate change and populations rise. Here in this chapter basics of Artificial Intelligence have been discussed how it can be related to agriculture.

1. Introduction

Artificial Intelligence has become most prominent in current generation. Every field is utilizing Artificial Intelligence for their better results and which also result in saving a lot of time. As Artificial intelligence is boosting different sectors to increase the productivity and efficiency and its utilization across different sectors like in the field of CIVIL Artificial Intelligence can be used to analyze the structures of buildings, dams etc., and Artificial Intelligence solutions are assisting in so many ways to overcome the traditional challenges in every field.

I. What is Artificial Intelligence

Artificial Intelligence is one of the most prominent field of Computer Science Engineering which attempts to redefine the tasks which are carried out by humans in their daily life with better results and maximum accuracy. Artificial Intelligence is actually meant to ease the work of human where the word artificial states the work done without any human interference and intelligence states the capabilities like human brain to perform tasks. The field artificial Intelligence has given so many definitions from time to time which can be typically stated as

Generation of new genetic variability through recombination breeding followed by appropriate selection procedure is one of the key factor for crop improvement. Gain in selection is restricted by correlated response of agronomically desirable as well as undesirable characters. There is a host of examples in various crop species where genes for desirable agronomic traits have been found to be closely linked with those governing undesirable characters (Murthy, 1971; Clegg et al., 1972). Such close linkages delay the realization of the full recombination potential in many crops. The biparental mating or the disruptive mating scheme, finds its importance in the improvement of autgamous speice,s because it increases the probabilities of obtaining useful recombinants as compared to selection without intermating (Hanson, 1959; Jensen, 1970).

In contrast to classical pedigree breeding, random mating population could produce large amounts of variability. This variability can be maintained in the population by the use of recurrent intermating, while the population is improved by selection. According to the standard breeding methodology in self-pollinated crops selection for quantitative characters are generally taken up in early segregating generations. For characters like yield, selection is continued till the material becomes homozygous. Such characters are controlled by large number of genes and a very large population has to be raised to make selection effective. Since most plant breeders could never grow the theoretically required size of population, the recombination is highly restricted. It is likely that many of the valuable genes may be practically lost in the process of advancing the material by continued selfing the generations. Andrus (1963) suggested crossing of selected sibs in early generation for reassembling the genes capable of functioning in a balanced polygenic system.

Among the different biometrical methods available, triple test cross analysis is one of the most efficient designs currently available for investigating the genetic structure of populations. It provides a test of epistasis and in its absence it gives independent and equally precise estimates of additive and dominance genetic components (Mather and Jinks, 1982).

11.1. Statistical analysis

Triple test cross

Jinks and Perkins (1970) method was applied to detect epistasis and estimate additive and dominance components of genetic variance.

The relative magnitude of different gene effects and an understanding of the mode of inheritance of complex quantitative traits have a direct bearing on the method of hybridization and selection which should be adopted in a specific breeding programme. Useful biometrical methods have been developed by the Birmingham group of scientists, to analyse the genetic architecture in organisms, through the use of first and second degree statistics obtainable from the true breeding inbreds, their first generation crosses and the different types of segregating progenies (Mather, 1949; Hayman, 1958; Gamble, 1962; Mather and Jinks, 1971). Since then this method has been widely used in the genetic analysis of various crops.

The narrow sense heritability estimates were high for days to first flowering, plant height at first flowering, number of leaves at first flowering, plant height at maturity, productive tillers per plant, boot leaf area, 100 grain weight, grain length and grain L/B ratio. These characters may be controlled by additive genetic variance. This indicated that the individual genotype can be evaluated readily from their phenotypic expression. Simple selection would more effective in the sets of materials exhibiting greater additive genetic variability and desirable mean performance. Thus, it merit selection in the next generation. On the other hand, the remaining traits may largely be controlled by non-additive genetic variance or the number of genes controlling these traits may be more, or these traits may largely be influenced by a large number of modifiers or might be largely influenced by environment. As the effects of the environmental variance were generally low, the observed low heritability for the above traits might be due other factors other than environment. These traits are hard to improve by selection. It necessitates progeny testing.

Eswaran (2007) made graphic analysis with the model proposed by Jinks and Hayman (1953). He evolved 30 hybrids from six parents in a diallel fashion. The results are presented in Tables 71-74

Eswaran (2007) made genetic analysis of six parents and their 30 hybrids of Okra with the method of analysis proposed by Hayman (1954b). He estimated the various genetic parameters viz., _D?, _F?, H?₁, H?² and Ê as well as the various genetic ratio. The results are presented in Tables 69 and 70.

The estimates of _D? were significant for seven out of ten characters studied. This indicated that the component of variation due to additive genetic variance was important for days to seedling emergence, number of first fruiting node, height of first fruiting node, plant height, days to first picking, number of fruits per plant and fruit weight. These characters could be improved by resorting to simple selection. The estimates of H? ₁and H? ₂were positive and significant for almost all the characters studied. It indicated that there were unequal frequency of alleles i.e., u ¹ v at all the loci, in this context, ‘u’ refers to the frequency of alleles which increase the mean expression of the character and are situated at loci which exhibited dominance. On the other hand, ‘v’ corresponds to the frequency of alleles at loci that decreases the expression of the character. Further proof for the unequal distribution of the alleles over loci was obtained

6.1. Line × tester analysis

A successful approach of crop breeding programme depends on the proper choice of best parents for hybridization and the ideal selection adopted in the early generation. But, the selection of parents for hybridization programme is relatively tough in the case of complex traits like yield and their components as they governed by large number of quantitative genes which are influenced by environments and seasons. Thus, breeders often meet with difficulty to fix in advance the desirable parents which will give superior progenies. In such a situation, knowledge on the nature of gene action on such complex quantitative traits of economic importance is necessary to plan and adopt appropriate selection techniques (Simmonds, 1979) and breeding methodology.

Combining ability analysis gives useful information regarding the selection of parents in terms of the performance of the hybrids. This analysis elucidates the nature and magnitude of various types of gene action involved in the expression of quantitative traits (Dhillon, 1975). Thus, the knowledge of combining ability serves as an useful tool for the selection of plants for hybridization and further exploitation. Combining ability consists of general and specific combining ability.

Sprague and Tatum (1942) defined general combining ability (gca) as the “average performance of a line in hybrid combinations” and specific combining ability (sca) “as those cases in which certain combinations do relatively better of worse than would be expected on the basis of the average performance of the lines involved. The general combining ability provides an indication of the importance of genes of largely additive in nature, while specific combining ability indicates the importance of non-additive gene effects.

Several biometrical techniques are available to estimate the general combining ability of parents and specific combining ability of hybrids. Among them, line × tester analysis is extensively used both in self and cross pollinated crops. The advantage of line × tester analysis is that a large number of inbred lines can be evaluated than diallel analysis. Besides, it provides additional information on the performance of the specific hybrids on gene action in contrast to top cross or polycross. The line × tester analysis is an efficient biometrical approach for assessing the combining ability of parents and hybrids (Kempthorne, 1957).

Genetic architecture of a population is the result of prolonged natural selection. The populations which exist in diverse environments might have been strongly diversified genetically. So, it is necessary to understand the extent and rates of genetic divergence existing between the diversified forms or types. Genetic diversity may be available in improved germplasm or it may exist only in genetically inferior stocks.

Genetic divergence among the parents is important in self pollinated crops because a cross involving genetically diverse parents is likely to produce high heterozygous as well as heterotic effect. In such a cross, more variability could be expected in the segregating generations due to the accumulation of the genes of both the parents. Genetic divergence, as one of the criteria of selection of parents is considered in plant breeding as early as 1962 (Murthy et al., 1962; Timothy, 1963). Later, Joshi and Dhawan (1966), Murthy and Arunachalam (1966), Arunachalam et al. (1984) and Lefort-Buson (1987) also stated that for exploiting heterosis as a means of increasing production, it is necessary to have parents of maximum divergence. The more diverse the parents, more is the chance of pronounced heterotic effects and increased spectrum of variability in the segregating generations.

The availability of statistical tools to quantitatively measure the genetic divergence between two or more populations and the relative contribution of individual characters to the total divergence have permitted to trace the evolutionary patterns in some crops such as rice and tobacco and in choosing the parents for hybridization in crop plants (Rao, 1958; Blackith, 1960; Morishima and Oka, 1960; Murthy et al., 1962). Among the several statistical methods developed for measuring the divergence between the populations, multivariate analysis (D² statistic) developed by Mahalanobis in 1936, has been found to be a potent tool. Mahalanbobis’s D² statistic is an effective tool in quantifying the degree of divergence at genetic level and provides a quantitative measure of the association between geographic and genetic diversity based on generalized distance. It is now known that the estimates of genetic divergence are also influenced by environments and seasons. Therefore, if the divergence analysis is conducted in a number of seasons and parents are selected on the basis of consistent divergence over the seasons, the problem of influence of seasons on divergence analysis can be largely overcome.

A successfully developed new cultivar should have stable performance and broad adaptation over a range of environments, seasons and locations, in addition to high yielding potential. Evaluating performance stability and range of adaptation is become increasingly important in breeding programmes. Since, stability and adaptability are important selection criteria in breeding programmes, in depth research on stability is needed for a better estimate of crop performance and adaptability.

Genotype × season interaction is of major performance to the plant breeder in developing stable but high yielding varieties of crop plants. When new varieties are tested over a series of seasons, the relative ranking of the varieties for any given attribute is rarely the same at each season. This increases the difficulty of identifying superior stable genotypes. Comstock and Moll (1963) have shown statistically the effect of large genotype × environment interaction in reducing the progress from selection.

In order to minimize the genotype × season interaction, stratification of the seasons can be practiced. The implication in stratification is that suitable selections could be made for unfavourable and favourable seasons encountered within the sampling units proposed. Stratification of the environments has been practiced for a long time to identify suitable selection that could be made for the narrower range of environments encountered within the sampling units proposed. However, even with this refinement of technique, the interactions of genotypes with locations in a sub-region with environments encountered at the same location in different years, frequently remain too large (Allard and Bradshaw, 1964). Little is known about the environmental factors contributing to such interactions. Sprague (1966) suggested that even if such information were available, the possibility of materially reducing the genotype × environment interaction in field experiments would remain questionable.

One of the important studies in the crop improvement programme is the discriminant function analysis. This is a method of exercising selection in plants, which show the extent of which character is genetically related to yield (Goulden, 1959). The technique of discriminant function was developed by Fisher (1936). But, the application of this method for plant selection was first made by Smith (1936) in wheat. He developed an index design for the selection of plant lines in wheat using the concept of discriminant function to derive a linear equation based on observable characteristics as the best available guide to the genetic value of each line.

Stephens (1942) reported about the interaction of yield components within the plant and within the population, which serves as an index for yielding ability. According to Hazel and Lush (1943), selection for a total score or net desirability is more efficient that selection for one trait at a time. The formulae presented by them for selection index give proper weight to each trait that are more efficient than selection for one trait at a time or for several traits with an independent culling level for each trait. According to Darlington and Mather (1949) discriminant function has been defined as a linear compound of series of varieties, obtained by giving the different varieties with individual coefficients which will minimize the difference between classes relative to variation within classes of objects of which the variates are measurements. Mather (1949) reported that these functions afford to the best available means of discriminating the classes and evaluating the genotype of a plant in terms of the observed values of various characters. Pundir and Raj (1971) reported that selection index based on three characters combination i.e., seeds per siliqua, 1000 seed weight and yield per plant was efficient in toria.

Seed yield is the foremost important attribute and is the end product of many complex component characters which singly or jointly influence the seed yield. Seed yield does not possess gene per se as such. It is the interaction phenotype of the harmonious understanding, mutual adjustment and manifestation of its component characters. As most of the component characters are polygenic, they are greatly influenced by the environment. So, the selection of genotypes based on yield as such is not likely to be effective. The information on the strength and direction of association of the component characters with seed yield and also among themselves will be very useful in formulating an effective breeding programme for seed yield improvement in a given time. Correlation studies provide about the magnitude and direction of genetic association between the seed yield and its component characters.

Correlation coefficients provide the measure of association between two characters. But, they do not provide the causal basis of such association. Further, correlation between seed yield and its component characters are often misleading due to the inter-relationship that exist among the component characters and the operation of size symmetry and component compensation that thwart the breeding progress. The direct contribution of each component to the seed yield and the indirect effects it has through its association with other components cannot be differentiated from mere correlation studies. This is how, it became necessary for a plant breeder to have information on the direct and indirect effects of these components on the seed yield to identify the key characters for improvement.