5 |
6 | ---
7 |
8 | ### **Web APP on Heroku**
9 | 
10 |
11 | **[The Project on Heroku](https://alcoholqualitychecker.herokuapp.com/)**
12 |
13 | ---
14 | ## The Dataset
15 | .png)
16 | ### **[Dataset](https://github.com/piyushpathak03/End-to-End-small-projects/blob/master/Alcohol-Quality-Checker/alcohol-quality-data.csv)**
17 | ---
18 | ## **Overview**
19 | * The Dataset has **'density'**, **'pH'**, **'sulphates'**, **'alcohol'**, **'Quality_Category'** columns. It has **4898 rows** and **5 columns**.
20 | * From the Dataset, we have to predict the **Quality of Alcohol**: **"High"** or **"Low"**.
21 | * **ExtraTreesClassifier** has been used for Feature Selection.
22 | * I have applied **Artificial Neural Network**, **Random Forest**, **Decision Tree**, **K-NN**, **Naive bayes classification** and **SVM** algorithms but at the end, **KNN** gave better results.
23 |
24 | ---
25 | ## **Machine Learning Pipelines:**
26 | ---
27 | ### **1> Feature Engineering:**
28 |
29 | **a> Handling Missing Values:**
30 | * Here, In this data, there is no requirement of handling missing values because already it is a complete dataset.
31 |
32 | **b> Feature Encoding:**
33 | * In this data, we do not have much categorical columns except output column.
34 |
35 | **c> Feature Scaling & Feature Transformation:**
36 | * For alogorithms like K-NN , K means , all neural network etc are based on some distance equations. Hence, they require Scaling.
37 | * But, when i applied it, there was not much of a difference in the accuracy so there was no meaning of using it. Because not everytime,we have to use scaling or transformation. * Transformation also seemed not required for this data because the distribution is almost gaussian for the required columns.
38 | ---
39 | ### **2> Feature Selection:**
40 | * There are various techniques for this but here i have used **ExtraTressClassifier**. For, this Project ExtraTressClassifier showed **2 columns** as most important **"sulfate"** and **"Alcohol Level"**.
41 |
42 | .png)
43 |
44 | ---
45 |
46 | ### **3,4&5> Model Selection**, **Model Creation**, **Testing**
47 |
48 | * To get the proper accuracy and for the proper splitting of the train and test data, I have used **Stratified K Fold Cross Validation** as it is very efficient in splitting the dataset.
49 |
50 | * Here, I have tried many algorithms like **Artificial Neural Network**, **Random Forest**, **Decision Tree**, **K-NN**, **Naive bayes classification** and **SVM**.
51 | * Among these, K-NN has gaven the higher accuracy (80%).
52 | * For this I have tried all the values k values **till 500** and **k=125** gave better results.
53 |
54 | | Algorithm | Average Accuracy |
55 | | ---- | ----|
56 | | Random Forest | 76.19% |
57 | | Decision Tree | 75.88% |
58 | | K-NN | 79.48% |
59 | | SVM | 79.7% |
60 | | Naive bayes | 78.21% |
61 | | ANN | 78.00% |
62 |
63 | ---
64 | * Finally, I decided to go with KNN because as we know **SVM generally has higher variance**, whereas in KNN we can fix it by **choosing the right K value**. In my project **k=125** gave better results.
65 | * For detailed look at Project, go to **[Alcohol-Quality.ipynb](https://github.com/piyushpathak03/End-to-End-small-projects/blob/master/Alcohol-Quality-Checker/alcohol-quality-data.csv)** and **[model.py](https://github.com/piyushpathak03/End-to-End-small-projects/blob/master/Alcohol-Quality-Checker/model.py)**
66 |
67 | ## About me
68 |
69 | **Piyush Pathak**
70 |
71 | [**PORTFOLIO**](https://anirudhrapathak3.wixsite.com/piyush)
72 |
73 | [**GITHUB**](https://github.com/piyushpathak03)
74 |
75 | [**BLOG**](https://medium.com/@piyushpathak03)
76 |
77 |
78 | # 📫 Follw me:
79 |
80 | [](https://www.linkedin.com/in/piyushpathak03/)
81 |
82 |
/My-Projects/Alcohole Quality/images/beer.jpg)
20 |
30 | 
6 |
7 | ---
8 |
9 | ## About Model:
10 | * The model is trained for classes: **'Goggles'**, **'Hat'**, **'Jacket'**, **'Shirt'**, **'Shoes'**, **'Shorts'**, **'T-Shirt'**, **'Trouser'**, **'Wallet'**,**'Watch'**.
11 | * Each class had around 30-50 images during training.
12 | * Images are trained with **VGG** Image Classification Model.
13 |
14 | 
15 |
16 | ---
17 |
18 | ## Understanding VGG:
19 |
20 | 
21 |
22 | * The full name of **VGG** is the **"Visual Geometry Group"**, which belongs to the Department of Science and Engineering of **Oxford University**.
23 |
24 | * In **[ILSVRC'14](http://www.image-net.org/challenges/LSVRC/#:~:text=The%20ImageNet%20Large%20Scale%20Visual,image%20classification%20at%20large%20scale.&text=Another%20motivation%20is%20to%20measure,indexing%20for%20retrieval%20and%20annotation.)**, VGG was **2nd in Image Classification** and **1st in Localization**.
25 |
26 | * The original purpose of VGG's research on the depth of convolutional networks is to understand how the depth of convolutional networks affects the accuracy and large-scale image classification and recognition.
27 |
28 | 
29 |
30 | * In order to deepen the number of network layers and to avoid too many parameters, a small 3x3 convolution kernel is used in all layers.
31 | * Input to VGG Model is **224x224** sized **RGB** Images, contains **3x3** and **1x1** filters and number of fully connected layers differs from VGG-11 to VGG-19.
32 | * 1x1 kernels is introduced to increase expressive power of network and reduce the amount of calculations without affecting input and output dimension.
33 | * In VGG, concept of **using multiple small kernels in multiple stacked Conv. layers** is used insead of **using less no. of Conv. layers and large kernels**, to reduce the model size by reducing total parameters. This adds **more non-linearity** as activation function (Relu) in used multiple times in a set of Conv layes.
34 | * Here, the receptive field of large kernel Conv. layer is same as stacked small kernels Conv. layers with **reduced parameteres**.
35 | * The overall structure includes **5 sets of Conv.** layers followed by **Max Pooling** layers.
36 |
37 | ---
38 |
39 |
40 |
41 | ## Implementing Project:
42 | 1. **Clone Repository and Install [Anaconda](https://docs.anaconda.com/anaconda/install/).**
43 |
44 | 2. **Create Conda Environment with Python 3.6:**
45 |
46 | conda create -n env_name python=3.6
47 |
48 | 3. **Install Libraries from Requirements.txt:**
49 |
50 | pip install -r requirements.text
51 |
52 | 4. **Run app.py:**
53 |
54 | python app.py
55 |
56 | This will run the app on your local machine.
57 |
58 | ---
59 |
60 | ## About me
61 |
62 | **Piyush Pathak**
63 |
64 | [**PORTFOLIO**](https://anirudhrapathak3.wixsite.com/piyush)
65 |
66 | [**GITHUB**](https://github.com/piyushpathak03)
67 |
68 | [**BLOG**](https://medium.com/@piyushpathak03)
69 |
70 |
71 | # 📫 Follw me:
72 |
73 | [](https://www.linkedin.com/in/piyushpathak03/)
74 |
75 |