Machine-Learning-Numpy

Code Machine learning models without any frameworks, Numpy only.

Table of contents

• Neural Network
• Clustering
• Decomposition
• Probabilistic
• Regression
• Trees based
• Timeseries
• Signal processing
• Monte-carlo

Neural Network

1. Deep Feed-forward

• gradient descent
• momentum
• nesterov
• rmsprop
• adagrad
• adam

2. Vanilla recurrent

• gradient descent
• momentum
• nesterov
• rmsprop
• adagrad
• adam

3. Long-short-term-memory recurrent

• gradient descent
• momentum
• nesterov
• rmsprop
• adagrad
• adam

4. gated-recurrent-unit recurrent

• gradient descent
• momentum
• nesterov
• rmsprop
• adagrad
• adam

5. Convolutional

• atrous 1D
• atrous 2D
• average pooling 1D
• average pooling 2D
• convolution 1D
• convolution 2D
• max pooling 1D
• max pooling 2D

6. batch-normalization
7. Dropout
8. Regularization
9. Neuro-evolution

• Iris classification
• Iris classification + Novelty search
• Regression

10. Evolution-strategy

Clustering

1. DBScan
2. K-Mean
3. K-Nearest Neighbors

Decomposition

1. Latent Dirichlet Allocation
2. Latent Semantic Analysis
3. Linear Decomposition Analysis
4. Non-negative Matrix Feature
5. Principal Component Analysis
6. TSNE

Probabilistic

1. Gaussian TF-IDF
2. Multinomial TF-IDF
3. Hidden Markov
4. Neural Network

Regression

1. Linear
2. Polynomial
3. Lasso
4. Ridge
5. Sigmoid logistic

Trees based

1. Decision Tree
2. Random Forest
3. Adaptive Boosting
4. Bagging
5. Gradient Boosting

Timeseries

1. Moving Average
2. Linear Weight Moving Average
3. John-Ehlers
4. Noise Removal-Get
5. Anchor Smoothing
6. Detect Outliers
7. ARIMA
8. Seasonal Decomposition

Signal processing

1. Convolutional 1D
2. Convolutional 2D
3. Pass-Filters

Monte-carlo

1. Markov Chain

• metropolis hasting normal distribution
• metropolis hasting stock forecasting

2. Pi estimation
3. Stock market prediction

Discussions

Some of results are not good because of softmax and cross entropy functions I code.

If found any error on my chain-rules, feel free to branch.