給創始人該想的問題

(從我一個尊敬的作家,賽斯·高汀

昨天有個朋友和我分享了一個新的商業想法。這讓我想到了一些商業模式的問題,在這裡用修辭性的疑問句來表達。如果您做好了,其他所有事情都會更容易:

您將如何發掘新的付費客戶?

為什麼您的付費客戶會告訴他們的朋友和同事你的存在?

這項業務會讓您既可以規模運作的實現又能讓你不斷的操作嗎?

這入門容易嗎?

如果是這樣,什麼會阻止其他人來開始做呢?

您將如何避免陷入低谷而陷入困境,像以中間人的身份來製造廉價商品?

隨時間操作後會變得更容易嗎?為什麼?

客戶為什麼必須選擇您而不是轉向其他更便宜或更方便的選擇?

能以廉價來創業,依靠廉價提供來有用的服務,並很大程度上是可替代的或隱形的企業通常會很難變成一項蓬勃發展的企業。對客戶的吸引力,網絡效應和情感聯繫是可以改變這一點,尤其是如果您能從一開始就將這些元素建立在這商業裡。

Image Regression Model for MNIST Handwritten Digits Using Python and AutoKeras

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 various machine learning algorithms and document the end-to-end steps using a template. The MNIST Handwritten Digits dataset is an image classification situation where we attempt to predict one of several (more than two) possible outcomes.

INTRODUCTION: The MNIST problem is a dataset developed by Yann LeCun, Corinna Cortes, and Christopher Burges for evaluating machine learning models on the handwritten digit classification problem. The dataset was constructed from many scanned document datasets available from the National Institute of Standards and Technology (NIST). Each image is a 28 by 28-pixel square (784 pixels total). A standard split of the dataset is used to evaluate and compare models, where 60,000 images are used to train a model, and a separate set of 10,000 images are used to test it. It is a digit recognition task, so there are ten classes (0 to 9) to predict.

ANALYSIS: Previously, we modeled the dataset using AutoKeras’ image classifier, and the system processed the validation dataset with an accuracy score of 94.84%. When we applied the best AutoKeras model to the previously unseen test dataset, we obtained an accuracy score of 98.4%.

After a series of modeling trials in this iteration, the AutoKeras’ image regressor system processed the test dataset with an RMSE score of 0.454 and an R2 score of 97.53%. When we applied the same predictions to the classification metrics, we obtained an accuracy score of 96.47%.

CONCLUSION: In this iteration, the best TensorFlow model generated by AutoKeras appeared to be suitable for modeling this dataset. We should consider experimenting with AutoKeras for further modeling.

Dataset Used: MNIST Handwritten Digits Dataset

Dataset ML Model: Image regression modeling with numerical attributes

Dataset Reference: https://www.tensorflow.org/datasets/catalog/mnist

One potential source of performance benchmark: https://machinelearningmastery.com/how-to-develop-a-convolutional-neural-network-from-scratch-for-mnist-handwritten-digit-classification/

The HTML formatted report can be found here on GitHub.

Image Classification Model for MNIST Handwritten Digits Using Python and AutoKeras

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 various machine learning algorithms and document the end-to-end steps using a template. The MNIST Handwritten Digits dataset is an image classification situation where we attempt to predict one of several (more than two) possible outcomes.

INTRODUCTION: The MNIST problem is a dataset developed by Yann LeCun, Corinna Cortes, and Christopher Burges for evaluating machine learning models on the handwritten digit classification problem. The dataset was constructed from many scanned document datasets available from the National Institute of Standards and Technology (NIST). Each image is a 28 by 28-pixel square (784 pixels total). A standard split of the dataset is used to evaluate and compare models, where 60,000 images are used to train a model, and a separate set of 10,000 images are used to test it. It is a digit recognition task, so there are ten classes (0 to 9) to predict.

ANALYSIS: After a series of modeling trials, the AutoKeras system processed the validation dataset with an accuracy score of 94.84%. When we applied the best AutoKeras model to the previously unseen test dataset, we obtained an accuracy score of 98.4%.

CONCLUSION: In this iteration, the best TensorFlow model generated by AutoKeras appeared to be suitable for modeling this dataset. We should consider experimenting with AutoKeras for further modeling.

Dataset Used: MNIST Handwritten Digits Dataset

Dataset ML Model: Image regression modeling with numerical attributes

Dataset Reference: https://www.tensorflow.org/datasets/catalog/mnist

One potential source of performance benchmark: https://machinelearningmastery.com/how-to-develop-a-convolutional-neural-network-from-scratch-for-mnist-handwritten-digit-classification/

The HTML formatted report can be found here on GitHub.

Time Series Model for University of Michigan Inflation Expectation Using Python and ARIMA

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

SUMMARY: This project aims to construct a time series prediction model and document the end-to-end steps using a template. The Inflation Expectation dataset from the University of Michigan is a time series situation where we are trying to forecast future outcomes based on past data points.

INTRODUCTION: The problem is forecasting the monthly number of median expected price change next 12 months based on consumers’ surveys. The dataset describes a time-series of percentages over 42 years (1978-2020), and there are 512 observations. We used the first 80% of the observations for training various models while holding back the remaining observations for validating the final model.

ANALYSIS: The baseline prediction (or persistence) for the dataset resulted in an RMSE of 0.221. After performing a grid search for the most optimal ARIMA parameters, the final ARIMA non-seasonal order was (4, 1, 2). Furthermore, the chosen model processed the validation data with an RMSE of 0.206, which was better than the baseline model as expected.

CONCLUSION: For this dataset, the chosen ARIMA model achieved a satisfactory result and should be considered for further modeling.

Dataset Used: University of Michigan: Inflation Expectation

Dataset ML Model: Time series forecast with numerical attribute

Dataset Reference: University of Michigan, University of Michigan: Inflation Expectation [MICH], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/MICH, October 24, 2020.

The HTML formatted report can be found here on GitHub.

Binary Classification Model for CycleGAN Horse vs. Zebra 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 various machine learning algorithms and document the end-to-end steps using a template. The CycleGAN Horse vs. Zebra dataset is a binary classification situation where we attempt to predict one of the two possible outcomes.

INTRODUCTION: The CycleGAN dataset collection contains datasets that consist of images from two classes A and B (for example, apple vs. orange, horses vs. zebras, and so on). The researchers used the images to train machine learning models for research work in the area of General Adversarial Networks.

In iteration Take1, we constructed and tuned machine learning models for this dataset using TensorFlow with a simple VGG-1 network. We also observed the best result that we could obtain using the test dataset.

In iteration Take2, we constructed and tuned machine learning models for this dataset using TensorFlow with a VGG-2 network. We also observed the best result that we could obtain using the test dataset.

This Take3 iteration will construct and tune machine learning models for this dataset using TensorFlow with a VGG-3 network. We will also observe the best result that we can obtain using the test dataset.

ANALYSIS: In iteration Take1, the baseline model’s (one layer with eight convolutional filters) performance achieved an accuracy score of 88.85% after 15 epochs using the unseen test dataset. After experimenting with different layer configurations, the best model (one layer with 64 convolutional filters) processed the validation dataset with 91.15% accuracy.

In iteration Take2, the baseline model’s (two layers with 8/16 convolutional filters) performance achieved an accuracy score of 90.00% after 15 epochs using the unseen test dataset. After experimenting with different layer configurations, the best model (two layers with 32/64 convolutional filters) processed the validation dataset with 91.92% accuracy.

In this Take3 iteration, the baseline model’s (three layers with 8|16|32 convolutional filters) performance achieved an accuracy score of 91.92% after 15 epochs using the unseen test dataset. After experimenting with different layer configurations, the best model (three layers with 48|96|192 convolutional filters) processed the test dataset with 96.54% accuracy.

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

Dataset Used: CycleGAN Horse vs. Zebra Dataset

Dataset ML Model: Binary classification with numerical attributes

Dataset Reference: https://people.eecs.berkeley.edu/%7Etaesung_park/CycleGAN/datasets/

One potential source of performance benchmarks: https://arxiv.org/abs/1703.10593 or https://junyanz.github.io/CycleGAN/

The HTML formatted report can be found here on GitHub.

Binary Classification Model for CycleGAN Horse vs. Zebra 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 various machine learning algorithms and document the end-to-end steps using a template. The CycleGAN Horse vs. Zebra dataset is a binary classification situation where we attempt to predict one of the two possible outcomes.

INTRODUCTION: The CycleGAN dataset collection contains datasets that consist of images from two classes A and B (for example, apple vs. orange, horses vs. zebras, and so on). The researchers used the images to train machine learning models for research work in the area of General Adversarial Networks.

In iteration Take1, we constructed and tuned machine learning models for this dataset using TensorFlow with a simple VGG-1 network. We also observed the best result that we could obtain using the test dataset.

This Take2 iteration will construct and tune machine learning models for this dataset using TensorFlow with a VGG-2 network. We will also observe the best result that we can obtain using the test dataset.

ANALYSIS: In iteration Take1, the baseline model’s (one layer with eight convolutional filters) performance achieved an accuracy score of 88.85% after 15 epochs using the unseen test dataset. After experimenting with different layer configurations, the best model (one layer with 64 convolutional filters) processed the validation dataset with 91.15% accuracy.

In this Take2 iteration, the baseline model’s (two layers with 8/16 convolutional filters) performance achieved an accuracy score of 90.00% after 15 epochs using the unseen test dataset. After experimenting with different layer configurations, the best model (two layers with 32/64 convolutional filters) processed the validation dataset with 91.92% accuracy.

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

Dataset Used: CycleGAN Horse vs. Zebra Dataset

Dataset ML Model: Binary classification with numerical attributes

Dataset Reference: https://people.eecs.berkeley.edu/%7Etaesung_park/CycleGAN/datasets/

One potential source of performance benchmarks: https://arxiv.org/abs/1703.10593 or https://junyanz.github.io/CycleGAN/

The HTML formatted report can be found here on GitHub.

Seth Godin’s Akimbo: The Zoom Revolution, Part 2

In his Akimbo podcast, Seth Godin teaches us how to adopt a posture of possibility, change the culture, and choose to make a difference. Here are my takeaways from the episode.

In this podcast, Seth discusses 17 changes that videoconferencing has influenced the way we work and live, compared to the traditional meetings. We need to be aware of the changes and perhaps even learn how to leverage the trend to our advantage.

Adding the ideas from numbers nine and ten, we will have number 11, which is transcription. Transcription means that we can automatically have a written record of everything that was discussed. We can go back over, figure out the intent, and make sure we are doing things right.

Number 12 is the idea of making meetings more computer aware. With our meetings now being digitized, we can now add more computing capabilities to our discussion. This idea is a big frontier because we could not find much use for getting computers involved in the traditional, analog-based meetings.

Number 13 is the idea of triggers. Video conferencing enables new changes and also make some old traditions optional. We are talking about a world where real estate is valued very differently. Commuting to the office for a meeting is becoming a choice, rather than mandatory. Meetings are getting shorter because people get called into discussions for the right reasons and asked to leave for the right reasons—many more triggers to come.

The number 14 is the idea of gamification. We play games with many things every day. Gamification is about status. Video conferencing allows us to gather measurements and statistics throughout the meeting. If we can figure out the proper metrics, why can we not make video conferencing a meaningful experience? Why would we not want to know that this will change the way we talk to each other professionally.

Number 16 is that it is always on. When we are in the office, the ability to have a meeting is confined to office hours. This idea of checking in with one another on an ad-hoc basis can boost productivity if used properly. There is nothing about computer video conferencing that makes that idea a difficult one.

Number 17 is the idea of scale and network effect. With the help of private enterprises and the Internet, video conferencing is free to get started. Once it is free, it is widely adopted. Once it is widely adopted, the network effect becomes ever more powerful. We are beginning to discover this significant impact on the way people are given a voice and how information flows.

Finally, number 18 is Chris Anderson’s idea of The Long Tail. Up until recently, we might only get to interact with a small group of people who are in our circle. Right now, we can interact with many more using video conferencing, and we have little in common with most of them. The long tail means we can get pockets of people who share something in common, regardless of their geographical locations. They do not even get a paycheck from the same company, but they desperately need to be connected.

This pandemic has brought a lot of damage, but one thing it has done is that it accelerated the arrival of the future. It moved the adoption of videoconferencing by three to five years ahead easily, and video conferencing will cause massive disruption in real estate, transport, and retail, to name a few areas.

If used properly, video conferencing can create so many opportunities for people to speak up, go outside their comfort zone, be heard, and be connected with the people they need to be associated with. We can certainly do more in figuring out how to create a platform for others to be great at it in this new frontier.

錯誤的選擇

(從我一個尊敬的作家,賽斯·高汀

當我們自由地做出個選擇,那個選擇會感覺就像是一個正確的行動。

但是,一旦我們意識到這是個錯誤, 我們通常不知道將來會學到什麼。

錯誤的選擇有可能是由於:

  • 信息上的差距
  • 劣質的分析(包括認知故障和對沉沒成本的依賴)
  • 來自同輩的壓力
  • 被操縱
  • 被喧囂
  • 權力上的不平衡
  • 太專注於短期
  • 被灌輸
  • 過於迷信
  • 未經審查的偏見

看一看:這些都是外來的產物,我們可以不學習也可以與它們隔離。這好消息是我們可以在處理選擇方面上來做得更好。