streamlit_test_app_21_11_2022 is a simple streamlit app in python for machine learning practice.

This app allows to see basic difference between Regression and Classification and some of their metrics. The pipeline was added to the app -the last part.

After selection of one file which content should be divided into train and test part, it should be checked and replaced the empty/NaN values (imputation). Secondly should be decided on the basis of choosen plots if the app will work with Regression or Classification and if any Scalling is needed before choose this.

For the Regression were used MAE, MSE and R2 metrics.

For the Classification were used classification report (which includes precision, recall and f1-score), confusion matrix, heatmap plots (plotly and matplotlib) and accuracy score which is included in classification report as well.

1. Because of personal selection of data, there is a possibility to give a title to your page as well (limit 100characters).
2. Then should be chosen a column which will be a target for a prediction. This decision should make easier a scatter matrix following by a bar plot and information how many classes the chosen target actually has. The limit for classification was reduced to 10 (That means in target can be maximal 10 classes). (Playing with data only recommended)
3. The table 'Number of missing values in columns' provides information if the data content NaN values and where these should be found - to make a right decision which imputation is appropriate to 'replace' them. They can be dropped at all or imputated according to their content (e.g. numerical column can not be imputated by 'constant text' and categorical by 'constant number' or 'mean'.) As the newest change there is a possibility to change every column. If positive or negative - depends on :)
4. As mentioned above Scalling should take part only in case of data with various value range. (Data with high value range will dominate calculations) ('Here should be possibility to choose only column/s and minmax Scaler as well later. Nowadays StandardScaler')
5. The next tasks depend on the data which should be processed but there is 'cross_val_score' or 'cross_validate' yet.

• Regression

: models available: LinearRegression, Lasso, SVR, RandomForestRegressor or XGBRegressor

• Classification

: models available: DecisionTreeClassifier, KNeighborsClassifier, RandomForestClassifier, SVC, LogisticRegression

Classification was separated to 'binary/multilabel' and 'multiclass' to be able to use specific metrics. ('SVC-Multilabel and samples do not work yet')

For some of the models is there an opportunity to change their parameters BUT in general all of these examples are simple.

(last update on the 1st of March 2023)