Author: David Lowe

Chief Technology Officer, Daines Analytics

Multi-Class Image Classification Deep Learning Model for Large Scale Fish Images Using TensorFlow Take 1

Template Credit: Adapted from a template made available by Dr. Jason Brownlee of Machine Learning Mastery.

SUMMARY: This project aims to construct a predictive model using a TensorFlow convolutional neural network (CNN) and document the end-to-end steps using a template. The Large Scale Fish Images dataset is a multi-class classification situation where we attempt to predict one of several (more than two) possible outcomes.

INTRODUCTION: This dataset contains nine different seafood types collected from a supermarket in Izmir, Turkey, for a university-industry collaboration project at Izmir University of Economics, and this work was published in ASYU 2020. For each class, there are 1000 augmented images and their pair-wise augmented ground truths.

In this Take1 iteration, we will construct a CNN model based on the InceptionV3 architecture to predict the leaf’s health state based on the available images.

ANALYSIS: In this Take1 iteration, the InceptionV3 model’s performance achieved an accuracy score of 99.65% after ten epochs using the training dataset. The same model processed the validation dataset with an accuracy score of 93.83%.

CONCLUSION: In this iteration, the InceptionV3-based CNN model appeared to be suitable for modeling this dataset. We should consider experimenting with TensorFlow for further modeling.

Dataset Used: A Large-Scale Dataset for Fish Segmentation and Classification

Dataset ML Model: Multi-class image classification with numerical attributes

Dataset Reference: Ulucan, Oguzhan and Karakaya, Diclehan and Turkan, Mehmet (2020), “A Large-Scale Dataset for Fish Segmentation and Classification,” 2020 Innovations in Intelligent Systems and Applications Conference (ASYU), IEEE (https://ieeexplore.ieee.org/abstract/document/9259867)

One potential source of performance benchmarks: https://www.kaggle.com/crowww/a-large-scale-fish-dataset

The HTML formatted report can be found here on GitHub.

Charlie Gilkey on Start Finishing, Part 8

In his book, Start Finishing: How to go from idea to done, Charlie Gilkey discusses how we can follow a nine-step method to convert an idea into a project and get the project done via a reality-based schedule.

These are some of my favorite concepts and takeaways from reading the book.

Chapter 8, Weave Your Project into Your Schedule

In this chapter, Charlie discusses the techniques to integrate our environments with the daily requirements demanded by our best work. He offers the following recommendations for us to think about:

  • Make sure our environment is working for us. Pay attention to these seven environmental factors and determine how they might affect us:
    • Sound
    • Smell
    • Sunlight
    • Clothing
    • Clutter/Organization
    • Amount of space
    • Music.
  • Batching and stacking are two techniques that can improve our efficiency.
    • Batching work is the process of doing similar kinds of work in a contiguous period.
    • Stacking work is the process of doing different but compatible kinds of work at the same time.
  • Keep the dread-to-work ratio down by dealing with the “Frogs.” Frogs are the tasks and chunks of projects that we do not want to do. However, we should consider addressing them early and often as necessary.
  • Focus more on the “when” rather than the “what.” Decide upfront when it is best to do particular work and stick to the schedule/plan.
  • First in priority doesn’t always mean first in the sequence. The key idea is to get to those high-priority tasks at the right time to complete the tasks in the most effective manner possible.
  • Use the 5/10/15 split to build daily momentum. We use our five projects to create and update our daily plan for ten minutes before we start and fifteen minutes at the end of our day.
  • Do not plan too far in advance. Doing so can create frustration and resignation because the further out we plan, the less likely our plan will be correct or practical.

忙碌（也可靠）

（從我一個尊敬的作家，賽斯·高汀）

讓您忙碌的事業很可能是您因可靠而贏得的聲譽。

但具有諷刺意味的是，這種非常的忙碌也可能會破壞您的聲譽。這就是為什麼許多服務提供商一旦開始受到關注，就開始做的跌跌撞撞的原因之一。

在這種危機來臨之前，您可以做兩件事：

首先，說很多的“不”。你在掙扎時所參加的事務，可能不是你現在應該參加的事務。您的可靠性聲譽為您贏得更多信任，而這種信任讓您受邀與更好的客戶合作，並開展更好的項目。這樣做的成本（收益）就是，您需要拒絕之前您願意接受的機會。

第二，說實話。這一開始很難做到，特別是因為我們的自我概念可能是建立在獨立性和無懈可擊的基礎上。但可靠性並不意味著完美性，這意思是說清楚實話。

這是可能有幫助的兩個座右銘：

“你會付出很多，但你得到的會比你付出的多了許多。”

和

“我們的秘訣是，我們不會為了想得到這個項目而撒謊。”

Algorithmic Trading Model for Mean-Reversion with RSI Exit Signal for a Group of Stocks Using Python

NOTE: This script is for learning purposes only and does not constitute a recommendation for buying or selling any stock mentioned in this script.

SUMMARY: This project aims to construct and test an algorithmic trading model and document the end-to-end steps using a template.

INTRODUCTION: This algorithmic trading model compares a simple mean-reversion strategy with or without using RSI as the exit signal for an individual stock. The model will use a trend window size of ten days for long trades only. When the 14-day RSI value reaches 70, the model will exit the long position.

ANALYSIS: In this modeling iteration, we analyzed 14 stocks between January 1, 2016, and July 16, 2021. The models’ performance appeared at the end of the script.

CONCLUSION: For all the stocks during the modeling time frame, the long-only trading strategy with or without RSI as the exit signal did not produce a better return than the buy-and-hold approach, except for NFLX. We should consider modeling these stocks further by experimenting with more variations of the strategy.

Dataset ML Model: Time series analysis with numerical attributes

Dataset Used: Quandl

The HTML formatted report can be found here on GitHub.

Binary Classification Image Classification Deep Learning Model for PlantaeK Jammu Kashmir Leaf Using TensorFlow Take 3

Template Credit: Adapted from a template made available by Dr. Jason Brownlee of Machine Learning Mastery.

SUMMARY: This project aims to construct a predictive model using a TensorFlow convolutional neural network (CNN) and document the end-to-end steps using a template. The PlantaeK Jammu Kashmir Leaf dataset is a binary classification situation where we attempt to predict one of the two possible outcomes.

INTRODUCTION: This dataset contains 2153 healthy and unhealthy plant leaf images for eight different fruits and vegetables. The plants taken for study are the native plants of the Kashmir region of India. Eight different plants, namely Apple, Apricot, Cherry, Cranberry, Grapes, Peach, Pear, and Walnut, are selected for the study based on their commercial and medicinal usage. The leaf is the primary object of reference taken for making the database, as they grow much earlier than fruits and the other plant parts.

In iteration Take1, we constructed a CNN model based on the InceptionV3 architecture to predict the leaf’s health state based on the available images.

In iteration Take2, we constructed a CNN model based on the DenseNet architecture to predict the leaf’s health state based on the available images.

In this Take3 iteration, we will construct a CNN model based on the Xception architecture to predict the leaf’s health state based on the available images.

ANALYSIS: In iteration Take1, the InceptionV3 model’s performance achieved an accuracy score of 95.07% after ten epochs using the training dataset. The same model processed the validation dataset with an accuracy score of 79.72%.

In iteration Take2, the DenseNet model’s performance achieved an accuracy score of 92.23% after ten epochs using the training dataset. The same model processed the validation dataset with an accuracy score of 75.06%.

In this Take3 iteration, the Xception model’s performance achieved an accuracy score of 93.45% after ten epochs using the training dataset. The same model processed the validation dataset with an accuracy score of 74.59%.

CONCLUSION: In this iteration, the Xception-based CNN model appeared to be suitable for modeling this dataset. We should consider experimenting with TensorFlow for further modeling.

Dataset Used: PlantaeK: A leaf database of native plants of Jammu and Kashmir.

Dataset ML Model: Multi-class image classification with numerical attributes

Dataset Reference: KOUR, VIPPON PREET; Arora, Sakshi (2019), “PlantaeK: A leaf database of native plants of Jammu and Kashmir,” Mendeley Data, V2, doi: 10.17632/t6j2h22jpx.2 (https://data.mendeley.com/datasets/t6j2h22jpx/2)

One potential source of performance benchmarks: https://data.mendeley.com/datasets/t6j2h22jpx/2

The HTML formatted report can be found here on GitHub.

Algorithmic Trading Model for Mean-Reversion with RSI Exit Signal for an Individual Stock Using Python

NOTE: This script is for learning purposes only and does not constitute a recommendation for buying or selling any stock mentioned in this script.

SUMMARY: This project aims to construct and test an algorithmic trading model and document the end-to-end steps using a template.

INTRODUCTION: This algorithmic trading model compares a simple mean-reversion strategy with or without using RSI as the exit signal for an individual stock. The model will use a trend window size of ten days for long trades only. When the 14-day RSI value reaches 70, the model will exit the long position.

ANALYSIS: In this modeling iteration, we analyzed the stock of AAPL (Apple Inc.) between January 1, 2016, and July 19, 2021. The mean-reversion model without using RSI produced a profit of 3.59 dollars per share, while the model with RSI signals returned 45.97. In addition, the buy-and-hold approach yielded a gain of 118.09 dollars per share.

CONCLUSION: For the AAPL stock during the modeling time frame, the long-only trading strategy with or without RSI as the exit signal did not produce a better return than the buy-and-hold approach. We should consider experimenting with more variations of the strategy.

Dataset ML Model: Time series analysis with numerical attributes

Dataset Used: Quandl

The HTML formatted report can be found here on GitHub.

Binary Classification Image Classification Deep Learning Model for PlantaeK Jammu Kashmir Leaf Using TensorFlow Take 2

Template Credit: Adapted from a template made available by Dr. Jason Brownlee of Machine Learning Mastery.

SUMMARY: This project aims to construct a predictive model using a TensorFlow convolutional neural network (CNN) and document the end-to-end steps using a template. The PlantaeK Jammu Kashmir Leaf dataset is a binary classification situation where we attempt to predict one of the two possible outcomes.

INTRODUCTION: This dataset contains 2153 healthy and unhealthy plant leaf images for eight different fruits and vegetables. The plants taken for study are the native plants of the Kashmir region of India. Eight different plants, namely Apple, Apricot, Cherry, Cranberry, Grapes, Peach, Pear, and Walnut, are selected for the study based on their commercial and medicinal usage. The leaf is the primary object of reference taken for making the database, as they grow much earlier than fruits and the other plant parts.

In iteration Take1, we constructed a CNN model based on the InceptionV3 architecture to predict the leaf’s health state based on the available images.

In this Take2 iteration, we will construct a CNN model based on the DenseNet architecture to predict the leaf’s health state based on the available images.

ANALYSIS: In iteration Take1, the InceptionV3 model’s performance achieved an accuracy score of 95.07% after ten epochs using the training dataset. The same model processed the validation dataset with an accuracy score of 79.72%.

In this Take2 iteration, the DenseNet model’s performance achieved an accuracy score of 92.23% after ten epochs using the training dataset. The same model processed the validation dataset with an accuracy score of 75.06%.

CONCLUSION: In this iteration, the DenseNet-based CNN model appeared to be suitable for modeling this dataset. We should consider experimenting with TensorFlow for further modeling.

Dataset Used: PlantaeK: A leaf database of native plants of Jammu and Kashmir.

Dataset ML Model: Multi-class image classification with numerical attributes

Dataset Reference: KOUR, VIPPON PREET; Arora, Sakshi (2019), “PlantaeK: A leaf database of native plants of Jammu and Kashmir,” Mendeley Data, V2, doi: 10.17632/t6j2h22jpx.2 (https://data.mendeley.com/datasets/t6j2h22jpx/2)

One potential source of performance benchmarks: https://data.mendeley.com/datasets/t6j2h22jpx/2

The HTML formatted report can be found here on GitHub.

Binary Classification Image Classification Deep Learning Model for PlantaeK Jammu Kashmir Leaf Using TensorFlow Take 1

Template Credit: Adapted from a template made available by Dr. Jason Brownlee of Machine Learning Mastery.

SUMMARY: This project aims to construct a predictive model using a TensorFlow convolutional neural network (CNN) and document the end-to-end steps using a template. The PlantaeK Jammu Kashmir Leaf dataset is a binary classification situation where we attempt to predict one of the two possible outcomes.

INTRODUCTION: This dataset contains 2153 healthy and unhealthy plant leaf images for eight different fruits and vegetables. The plants taken for study are the native plants of the Kashmir region of India. Eight different plants, namely Apple, Apricot, Cherry, Cranberry, Grapes, Peach, Pear, and Walnut, are selected for the study based on their commercial and medicinal usage. The leaf is the primary object of reference taken for making the database, as they grow much earlier than fruits and the other plant parts.

In this Take1 iteration, we will construct a CNN model based on the InceptionV3 architecture to predict the leaf’s health state based on the available images.

ANALYSIS: In this Take1 iteration, the InceptionV3 model’s performance achieved an accuracy score of 95.07% after five epochs using the training dataset. The same model processed the validation dataset with an accuracy score of 79.72%.

CONCLUSION: In this iteration, the InceptionV3-based CNN model appeared to be suitable for modeling this dataset. We should consider experimenting with TensorFlow for further modeling.

Dataset Used: PlantaeK: A leaf database of native plants of Jammu and Kashmir.

Dataset ML Model: Multi-class image classification with numerical attributes

Dataset Reference: KOUR, VIPPON PREET; Arora, Sakshi (2019), “PlantaeK: A leaf database of native plants of Jammu and Kashmir,” Mendeley Data, V2, doi: 10.17632/t6j2h22jpx.2 (https://data.mendeley.com/datasets/t6j2h22jpx/2)

One potential source of performance benchmarks: https://data.mendeley.com/datasets/t6j2h22jpx/2

The HTML formatted report can be found here on GitHub.