Today's Progress : Plan what to work on for first few weeks.
Thoughts : I'm looking forward to becoming more familiar with NumPy and gaining practice learning CNN's, RNN's, GAN's, and Deep Reinforcement Learning.
Today's Progress :
- Complete most of Topic 1 in my Learning Plan.
- Build a basic feedforward network with backpropagation using NumPy. This is based off of Trask's Neural Network in 11 Lines of Python
- Modify my basic feedforward network to handle dynamic number of layers.
Thoughts :
- It was a good refresher. I am feeling more comfortable with NumPy.
- I will move onto Topic 2 tomorrow and slowly continue reading the Deep Learning Book Part 1.
Link of Work:
- Commit of basic FeedForward Network - Fixed number of layers
- Commit of FeedForward Network - Dynamic number of layers
Today's Progress :
- Install Tensorflow on laptop
- Explore Tensorboard
- Read CS20 Tensorflow overview
- Read CS20 Tensorflow Operations
Thoughts :
- Tensorflow looks powerful. I'm looking forward to writing some test programs tomorrow and playing with Tensorboard more.
Link of Work:
- None today
Today's Progress :
- Read CS20 (Tensorflow for Deep Learning Research) lectures and slides
Thoughts :
- There is a lot of information in these slides. I will start to play with Tensorflow and then come back in a few weeks to understand them better.
Link of Work:
- None today
Today's Progress :
- Completed 2 tutorials on the Tensorflow website:
- Basic Classification Tutorial
- Text Classification Tutorial
Thoughts :
- It is very helpful to follow the tutorials. It helps me understand how to write Tensorflow and Keras code.
Link of Work:
Today's Progress :
- Finished Tensorflow Regression tutorial
- Researched hard drives, RAM, motherboard, and case for deep learning PC
Thoughts :
- Hoping to finish picking parts for deep learning PC and order them tomorrow.
Link of Work:
- Regression Tutorial Commit
Today's Progress :
- Picked out almost all of my Deep Learning computer's parts.
- Watched Siraj Raval's Reinforcement Learning for Stock Prediction Video
Thoughts :
- Hope to pick out the rest of the computer parts tomorrow.
Link of Work:
- None today
Today's Progress :
- Found all parts for Deep Learning PC. I was planning on using fans to cool, but realized I needed to do water cooling to sufficiently cool the 2 Nvidia Titan XP Collector's Edition graphics cards.
Thoughts :
- I will begin researching water cooling tomorrow.
Link of Work:
Today's Progress :
- Researched watercooling. Found other people that watercooled Titan XP Collector's Editions and began researching their parts.
- Found EK Fluid Gaming parts
- Found an interesting article saying that evolutionary algorithms are able to outperform deep-learning at video games
Thoughts :
- The concept of watercooling is simple, I just need to determine exactly which parts I need to sufficiently cool my PC.
- I will write a simple evolution algorithm tomorrow
Link of Work:
- None today
Today's Progress :
- Completed a genetic algorithm to solve the Traveling Salesman Problem.
- Emailed EK Fluid Gaming's help desk to find out what components I needed to cool my Deep Learning PC. After talking with them, I determined I needed:
- A360G kit (I won't use the pump that comes with it though)
- D5 pump
- EK-AC GEFORCE GTX Water Block
- Vertical pump mount (to mount the pump combo in my case to a fan or radiator so I don't have ot drill holes)
- Multi-GPU kit
- EK-FG 360 Expansion Pack to get a second radiator (one radiator would not sufficiently cool the CPU and 2 GPUs)
- Colored fluid (if I want it). The kit comes with clear fluid
Thoughts :
- The genetic algorithm was interesting to implement. I plan to revisit this more after I do some reinforcement learning to compare the results.
- The help desk at EK Fluid Gaming was very helpful and I am understanding better the parts needed and what they do.
- I need to find a case that will fit my motherboard (it is SSI CEB form factor... I may try to find a different motherboard as well), 2 radiators, and all of the watercooling components.
Link of Work:
- Genetic Algorithm - Traveling Salesman Problem Commit
Today's Progress :
- My Jupyter Notebook stopped working, so I had to troubleshoot and ended up re-installing it.
- Finished Tensorflow tutorial on Overfitting and Underfitting
- Watched YouTube videos by Steve Brunton on using Genetic Programming to optimize the structure and parameters of an effective control law
Thoughts :
- I plan to finish the Tensorflow tutorials this weekend, and start something larger before the end of the weekend.
- I want to play with Genetic Algorithms more on future projects. I find them really interesting.
- I also plan to finalize my computer parts (hopefully finding a computer case tomorrow) and order the parts before Monday.
Link of Work:
- Tensorflow Tutorial - Overfitting and Underfitting Commit
Today's Progress :
- Finished the Tensorflow tutorial on saving and loading models.
- Watched Siraj Raval's Intro to AI for Video Games (Reinforcement Learning)
- Found a great overview of Reinforcement Learning concepts by Emmanuel Ameisen
- Emmanuel linked to an amazing set of tutorials by Arthur Juliani
Thoughts :
- I look forward to going through Emmanuel's and Arthur's articles in more detail and understand Reinforcement Learning at a deeper level.
Link of Work:
- Tensorflow - Save and Restore Models Commit