Home About LinkedIn Google Scholar
About AI Topics Data Science Topics Programming Topics System Design Topics (Book) Kubernetes Coding Examples Machine Learning RAG Concept RAG Implementation REDIS Benchmarking Code Llama Intro Code Llama Hands-On Breaking Code Llama Why Jhonny Cant't Prompt Discriminative Vs Generative Models Generative Adversarial Models Regex in python Kube Health Checks Raft Algorithm (Consensus) Ex-Google SRE on Alerting System Design Primer (Github) Supervised Machine Learning Unsupervised Machine Learning (Clustering) Unsupervised Machine Learning (Dimensn Redcsn) Unsupervised Machine Learning (Timeseries) Deep Learning (Introduction) PostgreSQL Natural Language Processing

Unsupervised Learning

Timeseries Analysis

Autocorrelation Plot

TODO

Decomposing Time series

in python use, from statsmodels.tsa.seasonal import seasonal_decompose . The output gives out all the three components, trend, seasonal and residuals. We can specify additive or multiplicative as the model to be used. The freq or the period argument is an important hyperparameter.

Stationarity

ARIMA and SARIMA

from statsmodels.tsa.statspace.sarimax import SARIMAX srma = sm.tsa.statesace.SARIMAX(data[temp], order=(p,d,q), seasonal_order=(0,1,1,12), trend='c').fit() srma.summary() srm.predict(start=start_idx,end=end_idx) # predict values for given index range

The AIC and BIc values printed from the summary() above should be as little as possible

There are auto fit models that will try to find the best models and bit the data. For example, for ARIMA, there is auto_arima .

Facebook prophet

It adds g(t), s(t), h(t). Its an additive model. g(t) models the non-periodic functions, s(t) models the seasonality and h(t) models the holidays. we would prefer historical data of at least a year preferrably to make a good prediction.

from fbprophet import Prophet model = Prophet() #you make want to take a log of data before fit() to make the variance smaller model.fit(my_data) #let us make a prediction for a year, by supplying 365 (no of days) #.. this will create an empty dataframe with just the dates forecast_dates = model.make_future_dataframe(periods=365) forecast = model.predict(future)

To work with holidays, for the train data, you must specify the holidays in a dataframe; and supply this df as argument when creating Prophet object, for example, model=Prophet(holidays=holidays)

thanksgiving = pd.DataFrame({ 'holiday': 'thanksgiving', 'ds': pd.to_datetime(['xxxx-xx-xx','xxxx-xx-xx']), 'lower_window': 0, 'upper_window': 1, }) # Similarly, create more if you want to holidays = pd.concat((thanksgiving,another_holiday)) model = Prophet(holidays=holidays) forecast = model.fit(my_data).predict(forecast_dates)

You can also add regressor using add_regressor , and do a cross validation using .cross_validation()

Deep Learning

RNN: Recurrent Neural Networks

Coding examples for RNN:

import tensorflow as tf from tensorflow import keras from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, SimpleRNN, LSTM, Activation, Dropout model = Sequential() noofhiddenunits = 10 #could be any number model.add(SimpleRNN(noofhiddenunits,input_shape=xxx)) model.add(Dense(1)) model.compile(loss='mean_squared_error',optimizer='adam') model.fit(x_train,y_train, epochs=epoch_num, batch_size=64) #to print the detail of the model model.summary()

Similar is the code example for LSTM:

import tensorflow as tf from tensorflow import keras from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, SimpleRNN, LSTM, Activation, Dropout model = Sequential() noofhiddenunits = 10 #could be any number model.add(LSTM(noofhiddenunits,input_shape=xxx)) model.add(Dense(1)) model.compile(loss='mean_squared_error',optimizer='adam') model.fit(x_train,y_train, epochs=epoch_num, batch_size=64) #to print the detail of the model model.summary()

Survival Analysis