Semester 1 Flashcards
Accuracy?
(TP+TN)/(TP+TN+FP+FN)
Bayes Theorem
P(A/B)=P(A)*P(B/A)/P(B)
Posterior=Prior*Likelihood/Evidence
Bayesian vs frequentist statitics
Bayesian asumes that the data are true, and that the parameters don’t have a single value, but a range of plausible values
Bernoulli distribution?
Single trial, only 2 possible outcomes.
Convolutional Network?
Decision Tree?
DSFI?
Error due to bias?
Model doesn’t learn the patterns of the training data. It oversimplifies the patterns or learns the wrong patterns
Error due to variance?
Model pays too much attention to patterns in the training data and “memorizes” the patterns. The model isn’t flexible enough to generalize into new data
Gamma distribution?
Continuous probability distribution that models right-skewed data
Gradient Boosting Machine
How to measure dependence between variables if the variables are continuous? Discrete?
- Pearson.
- Spearman.
KNN?
K Nearest Neighbours. Classification or Regression. Finds the K nearest points to a given new data point and uses the majority or average vote to predict the target variable.
Likelihood estimation method?
What is the likelihood of the variable having a certain distribution, given the observed data?
Precision?
TP/(TP+FP)
Medida de qué tan hacia los extremos está una cierta distribución?
Kurtosis
Recall?
TP/(TP+FN)
Skewness?
Si la distribución está cargada hacia la izquierda o la derecha.
Tipos de Naive Bayes?
Gaussian: continuous variables
Multinomial: discrete variables
Binary: binary variables
Valores de distribución Normal, para +-1d.e? +-2d.e? +-3d.e?
34.1% (68.2%)
13.6% (95.4%)
2.1% (99.7%)
Poisson distribution?
Models the probability of an event happening k times whithin a given interval of time or space.
False Positive Rate?
1-Specificity = FP/(FP+TN)
True Positive Rate?
TP/(TP+FN)
Specificity?
TN/(TN+FP)
Formula de Probabilidades? Odds? Como pasar de Odds a Probabilidades?
a. F/(F+NF).
b. F/NF.
c. b/b+1
Neural Network?
Input layers + hidden layers + output layers.
Each neuron has a weight for each past neuron + one bias + function (usually sigmoid).
It calculates a Cost function and then looks for the negative gradient, which tells us which weights and biases we need to change in order to lower the cost, this is called Backpropagation.
Stochastic gradient descent?
Stochastic means doing the gradient descent process but not with all samples at the same time, instead you compute it for one batch, take a step and then compute it for the next batch, and take another step, etc.