Trying to implement some of machine learning algorithms and techniques on C++. The aim of doing this is not replacing python with C++ but to learn C++ for practical purposes.

To run my current progress, compile and run.cpp as follows:

g++ run.cpp -std=c++17 -o run.o
./run.o

I am using std::any for which C++17 is required, hence the -std=c++17 flag.

bash run.sh

^faster. TODO: MAKEFILE.

• A simple CSV reader. See CSVReader.h.
• A label encoder for categorical variables. See preprocess.h.
• Summarizer in preprocess.h. Prints "mean", "stdv", "count", "min", "20%", "40%", "50%", "60%", "80%", "max" for each numerical column.
• Summarizer updated in preprocess.h. Prints "count", "unique", "mostfrq", "lstfrq" for each categorical column.
• FillNA added in preprocess.h. Supports mean (mostfreq for categorical) and zero value filling for all columns.
• Feature engineering - Added pearson correlation for all pairs of numcols in featureengine.h.
• Data slicing: (See CSVReader.h)
  • Supports negative index (-i means i indices from end)
  • Asserts for ranges
  • Inplace update for saving memory
  • Default arguments for start and end index of both rows and cols

• Implemented some loss functions (losses.h):
  • MSE
  • RMSE
  • MAE
  • Huber
• Implemented some metrics:
  • Accuracy
  • Multi label confusion matrix
  • Precision
  • Recall
  • F1
  • Fbeta

• A versatile feature engineering engine.
• Image data - resize, crop, negative, blur, etc.

• Line plot
• Bar plot
• Point plot

• Cross entropy, negative log loss, Hinge
• KL Divergence

• KNN
• Linear regression
• Polynomial regression
• Logistic regression
• Kmeans
• PCA
• GMM
• tSNE
• MLP
• Random forests
• Gradient engine??!
• DNNs??