Today's Progress: Watched youtube video by Josh Gordon on how to build decision tree classifier from scratch.
Thoughts: In decision tree classifier you basically take all data with labels and put it in root node and ask question on it. Then based on true or false answer you split the data into two chunks which become next nodes of the tree. This goes on untill there is no further questions left to ask. Note: This is perhaps the worst way to descript it. lol. I'll keep improving.
Link to work: https://www.youtube.com/watch?v=LDRbO9a6XPU
Today's Progress: Studied the algorithm in detail and started writing the program in javascript
Thoughts: I have written function to find unique values from a column in dataset. Format of my data is different than that mentioned in tutorial so I will have to improvise for the new format. I am studying from here: https://github.com/random-forests/tutorials/blob/master/decision_tree.ipynb
Link to work: https://github.com/VariSingh/Decision-tree-classifier-from-scratch
Today's Progress: Studied the decision tree algorithm in details. Added function to get unique values from labels
Thoughts: I studied Gini impurity and information gain as well. It seems like I now understand these things in isolation but I need to study it more before I can see the big picture of this classifier and write all code out of my head.
Link to work: https://github.com/VariSingh/Decision-tree-classifier-from-scratch/commit/6ce7262582a22de56dd6fe016da485b6b2c64638
Today's Progress: Studied how array can be split by finding gini impurity and information gain.
Thoughts: Didn't commit anything to repo. Studied find_best_split() function from below link. I couldn't get the enough workable version to commit.
Link to work: https://drive.google.com/file/d/1gURSnd1VjlaVnbh6rrObirnWlVplOnRZ/view?usp=sharing
Today's Progress: Worked on function to split data based on best information gain.
Thoughts: I had to watch the youtube video by Josh Gorden multiple times before I could understand what the code is doing :p. I feel like I am crawling but finally understood how this works. Added function to split data based on gini impurity and best information gain. The function is not complete yet.
Link to work: https://github.com/VariSingh/Decision-tree-classifier-from-scratch/commit/5a8e01dcf583c14d8e5c3580fb2c76efc547f187
Today's Progress: Updated function to split data and ask question.
Thoughts: We pick one value from a feature and apply for loop over it, then compare(ask question) it with every other value of the feature. Those which return true move to right branch. Others go to left branch. I have not tested the code yet. I feel like I am working much slower than expected. Need to speed up.
Link to work: https://github.com/VariSingh/Decision-tree-classifier-from-scratch/commit/e3fbd5a8e37e70c345583b8188584ab5dcb60e0b
Today's Progress: Tested functions and fixed bugs.
Thoughts: I added index.html file so that I could debug functions. Now row splitting and gini impurity is working. It seems like keeping features and labels in separate arrays is not a good idea. I will keep it like that though.
Link to work: https://github.com/VariSingh/Decision-tree-classifier-from-scratch/commit/6191bc0c8de9be904a7bdb995bdc2f4fdfbd1801
Today's Progress: Fixed bug in information gain, update in best_split function.
Thoughts: Not much to explain today. Just debugging and bug fixing. I feel like I am spending more time on coding aspects of ML and not really logic thing. Logic is something that goes more in head and less on keyboard. I will spend some time thinking about how to make this better.
Link to work: https://github.com/VariSingh/Decision-tree-classifier-from-scratch/commit/453dfd3789abbffa5a5bacdefc8ad33aa4b1cdae
Today's Progress: Added recursive call to classifier function. Now tree is split on each node.
Thoughts: When I run function recursively it gives some weird looking response which apparently means I am doing something wrong. I will fix this in next commit.
Link to work: https://github.com/VariSingh/Decision-tree-classifier-from-scratch/commit/c35eb726b44b3e9a8dd4e11e0f3c0db47e99b0d4
Today's Progress: Learnt about SVM(Support Vector Machine) from Udacity course "Intro to machine learning".
Thoughts: I have put decision tree classifier aside for some time because I felt like I am stuck with one algorithm only. I will come back to it later. So I read about SVMs, kernel trick, how to convert linear SVMs to linear. I have completed 22 chapters from the course on SVMs. Link at the bottom.
Link to work: https://in.udacity.com/course/intro-to-machine-learning--ud120-india
Today's Progress: Read about Naive Bayes Classifier from Udacity course "Intro to machine learning".
Thoughts: I couldn't understand most part of it. Even though I watched most of the videos, I seriously didn't get anything from the theory part (Bays Rule Diagram or Cancer test quiz). Really disappointed.
Link to work: https://in.udacity.com/course/intro-to-machine-learning--ud120-india
Today's Progress: Read about probability, Bayes Theorem.
Thoughts: I found out that I can't proceed with SVM's practical samples on Udacity because a part of it is in Naive Bayes lecture and I couldn't do that because I couldn't understand the theory part. So I jumped to very basics - probability theory. I am studying probability from Khan Academy.
Link to work: https://www.khanacademy.org/math/statistics-probability
Today's Progress: Probability continue.
Thoughts: Still reading probability from Khan Academy. Watched more lectures in order to understand it better.
Link to work: https://www.khanacademy.org/partner-content/wi-phi/wiphi-critical-thinking/wiphi-fundamentals/v/bayes-theorem
Today's Progress: Studied addition rule of probability, sets.
Thoughts: It seems like I'm back into school and studying things that I thought I would never study again but this is interesting. I need to study this anyway so that I understand fundamentals.
Link to work: https://www.khanacademy.org/math/statistics-probability/probability-library
Today's Progress: Studied multiplication rule for dependent and independent events.
Thoughts: Doing lot of theory. I have the urge to actually code but I won't really understand what I am doing untill I know how this stuff works. One aspect of that is probability. Udacity theory videos by Sebastian Thrun are really hard to understand until I know basics of probability.
Link to work: https://www.khanacademy.org/math/statistics-probability/probability-library
Today's Progress: Studied Naive Bayes theory from Udacity course on machine learning by Sebastian Thrun.
Thoughts: In order to understand this course you must learn "Intro to statistics". I read basics of probability from KhanAcademy. Now I am on a mini project at the end of the "Naive Bayes" lesson.
Link to work: https://in.udacity.com/course/intro-to-machine-learning--ud120-india
Today's Progress: Worked on mini project (Naive Bayes)but could not get it working because of python version issue.
Thoughts: Udacity has good content but many times I feel completely lost. Either it is intentional or they presume that I know lots of maths and already a wizard :p. e.g I also started reading deep learning course and the first thing they ask is to write a softmax function. How the hell can I write it before knowing it?:D Perhaps I can't skip through the lectures.
Link to work: https://github.com/udacity/ud120-projects
Today's Progress: Watched youtube video by @sentdex on Deep neural networks.
Thoughts: I have already done one program on Neural network but when I saw the video I was like why I didn't see this before :D. I'm going to implement it the next thing.
Link to work: https://www.youtube.com/watch?v=PwAGxqrXSCs&list=PLQVvvaa0QuDfKTOs3Keq_kaG2P55YRn5v&index=47
Today's Progress: Started working on fashionMNIST with deep neural network.
Thoughts: I played with the data a little bit. Couldn't plot the data. I was looking for something that Tensorflow can do for me. I can use matplotlib also but I need to study it more.
Link to work: https://github.com/VariSingh/deepMNIST
Today's Progress: Wrote code to create 4 layers for deepMNIST.
Thoughts: I did not get much time today so watched the video by @sentdex on Deep neural network and followed the same pattern to write 4 layers on neural network.
Link to work: https://github.com/VariSingh/deepMNIST/commit/268ffd07d0596b5b54beca280c281f85a65b372f
Today's Progress: Wrote code to calculate weighted sum of input and bias.
Thoughts: Wrote four variables which have inputs * weights + biases. Also studied tf.random_normal. tf.zeros, difference between session and interactive session. I need to study python as well because I am not good at it.
Link to work: https://github.com/VariSingh/deepMNIST/commit/6cad248752d3f57200d6fa3fbe5a1a8298ae65a5
Today's Progress: Finished deepMNIST.
Thoughts: I think this can be tweaked in more than one way to affect the accuracy. I will play with it a little more tomorrow.
Link to work: https://github.com/VariSingh/deepMNIST/commit/b1107d203d80bed4b0146066b5c415c038e9abea
Today's Progress: Watched @sentdex youtube video on using raw text data for sentiment analysis.
Thoughts: This looks more like a practical thing you will face in real life where your data is not properly labeled.
Link to work: https://github.com/VariSingh/sentiment_analysis/commit/3c402e634d99bcd7795a1c7e110583fa09ed430c
Today's Progress: Continued with sentiment analysis code.
Thoughts: Watched the same video again and continued to code. Basically imported WordNetLemmatizer and tried it. Didn't give me any impressive results so I pushed to repo whatever was done. While writing code I realized I don't know basic python so moved to w3schools. I know it's funny. Here's the link to last chapter I read - https://www.w3schools.com/python/python_pip.asp
Link to work: https://github.com/VariSingh/sentiment_analysis/commit/e5fa7e37aa7eb43a2237ca15761b265f4e38337c
Today's Progress: Continued with sentiment analysis code for third day.
Thoughts: Watching the @sentdex video and coding along. I don't know much of python so sometimes I have to spend more time trying to figure out what that piece of code is doing. Learnt again list,tuple,set. Read about Counter module.
Link to work: https://github.com/VariSingh/sentiment_analysis/commit/3b54959632dc4c7450f6997b9d36f0e04e9fbe86
Today's Progress: Sentiment analysis - Wrote code to generate file with features and labels. Fixed bugs.
Thoughts: I continued with same program today. Fixed some bugs. All bugs are not fixed yet. Need to study more about python. I also studied about convolutional neural networks.
Link to work: https://github.com/VariSingh/sentiment_analysis/commit/2a49d2174eecf27f398307c84e0977737b5c94e3
Today's Progress: Read about tflearn and Keras.
Thoughts: Today I did not cotinue with sentiment analysis because I am stuck in a python bug. I watched Youtube video by Siraj Raval on sentiment analysis. I want to decrease the time it takes to learn and implement DNN that's why I explored tflearn. Found a Reddit link as well.
Link to work: https://www.reddit.com/r/MachineLearning/comments/50eokb/which_one_should_i_choose_keras_tensorlayer/
Today's Progress: Studied create_lexicon function in detail.
Thoughts: Although I have coded a part of the sentiment analysis by watching sentdex youtube video but there are lots of things that I really don't understand yet especially on python part. I run one function at a time and add a print statement to understand each line.
Link to work: https://github.com/VariSingh/sentiment_analysis
Today's Progress: Studied sample_handling function in sentiment analysis.
Thoughts: It's funny, I thought I understood whole program. Two days later I am debugging whole application and learning new things from it. In sample_handling we are converting text to vector form which can be fed into tensorflow.
Link to work: https://github.com/VariSingh/sentiment_analysis
Today's Progress: Fixed bug in featureset list in sentiment_analysis.
Thoughts: I think it was because of the different version of python I have. I don't know but the code from sentdex video didn't work for me. I had to create a sample list and experiment on it with some stackoverflow help finally fixed the problem. Will work on Neural network tomorrow.
Link to work: https://github.com/VariSingh/sentiment_analysis/commit/90b160a5eb321e16fbc0fea477dec5f7de8b530d
Today's Progress: Trained and tested sentiment_analysis.
Thoughts: modified existing neural network to take data that I generated yesterday and trained and tested it. Also read about basics of Recurrent Neural networks and Convolutional Neural Network.
Link to work: https://github.com/VariSingh/sentiment_analysis/commit/920f931875339fb840599bac7a73871e7cc534f4
Today's Progress: Studied Convolutional Neural Network.
Thoughts: I watched multiple videos on Convolutional Neural networks which included concept and coding. I have created repository for this but haven't coded anything yet.
Link to work: https://github.com/VariSingh/MNIST_with_Convnet
Today's Progress: Studied ConvNet in more detail.
Thoughts: I watched more videos on ConvNet. I am still not having a clear picture of how the convolution works. I also watched a video on how Generative Adversarial Network work.
Link to work: https://www.youtube.com/watch?v=mynJtLhhcXk&index=57&list=PLQVvvaa0QuDfKTOs3Keq_kaG2P55YRn5v
Today's Progress: Studied basics of multilayer perceptron.
Thoughts: While studying ConvNet I realized I do not know some pieces of Vanilla Neural Network. So I watched the videos by 3blue1brown to understand cost function and optimization. I am still not very sure about that. I also studied tflearn and kind of compared it with plain tensorflow. I think I should create more samples of multilayer perceptron with tflearn.
Link to work: https://www.youtube.com/watch?v=IHZwWFHWa-w
Today's Progress: Studied ConvNets.
Thoughts: I learnt what is filter, stride, size. I also studied about tflearn but still I need to learn that again so that this all goes in my subconcious mind.
Link to work: https://www.youtube.com/watch?v=SQ67NBCLV98&t=16s
Today's Progress: Watched video about GAN.
Thoughts: I watched Siraj Raval's video about Generative Adversarial Network.
Link to work: https://www.youtube.com/watch?v=0VPQHbMvGzg
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent