zilutian / ml_projects Goto Github PK

f1 score

Apply method to data. Include

• cross-validation to choose hyper parameter gamma;
• analysis of size of neighbourhood k.

Classification

Once data matrix is built. Do the following:

1. Feature selection (PCA???)
2. Clean data (??)

• Stochastic gradient
• Logistic regression

alternative:

• average
• median
• some prediction

Figure out a strategy for building features for image data.

• RGB colors
• Histogram of Oriented Gradients
• Local Binary Pattern
• Other...

(Check Scikit image module)

1. Missing data
   Remove missing data for this analysis:
2. Statistical characteristics of features (mean, std), types of variables
3. Check each feature 1 degree regression
4. Correlation with output
5. Cross correlation with each other

Get a sense of what the data set looks like

Adam optimizer, etc, instead of momentum optimizer

• Normalization (average/mean)
• Remove outliers (-999)

Normalize data (Global and local)