model-logger

model-logger is a Python library for storing model's profile and rapid inter model comparision. Powered by dash and SQLITE3, It's compact ,light weight ,interactive yet powerful tool to gain useful insights.

Installation

Using pip

Use the package manager pip to install model-loger.

Announcements

Version v0.2.3 released

• Support for Catboost
• Optional Test score added (One of the two scores is compulsory)
• Test vs Validation Score across models comparision graph added
• Best model functionality added.

pip install modellogger

Initialization

from modellogger.modellogger import ModelLogger

#initialise a modelloger instance
path = "c/path/to/db/databasename.db"
mlog = ModelLogger(path) #setup complete
 

If you are already using a db created by modelloger you can directly load it by stating it's path If you are creating a new project just give location where you want to store the db followed by a name.db .

• [Try It Now?](cause where is the fun in reading documentation XD) [](https://colab.research.google.com/github/SohamPathak/modellogger.github.io/blob/master/assets/sample/model-logger .ipynb)

Now you are ready to rock and roll. Out of the box you will have the following functionalities:

Usage

store_model()

save_pickle : let's you save the model as a pickle file with model_name as the file name . Uses joblib for pickling ,to use it later use joblib.load('name').

model_name : Give a unique name to your model.

columns :Columns used for tranning the model.

accuracy : Scores to measure the performance eg. rmse , mse , logloss or a custom function that returns a metric. Ideally same factor across all models will help gaining insights from summary.

Flag : If true than will print out the contents of the db.

model-logger currently stores the following attribute:

• Model id
• Model name(given by user)
• Scores
• Number of columns
• Feature set used for tranning the model
• model parameters
• Train size
• Number of continuous variables
• Number of catagorical variable
• Model type
• Model name

Example

 gboost = GradientBoostingClassifier() 
 gboost.fit(X_train,y_train) 
 mlog.store_model('my_model_name',gboost,X_train,1.0) 
 #alternatively
 mlog.store_model('my_model_name',gboost,X_train,get_score(gboost),save_pickle = True)

view_results()

generate_csv: If true than generate the report in the form of a csv

csv_name: Name of the csv file , default -- Model_Report.csv

Example

mlog.view_results(True,'my_report')

delete_model()

Model_name : name of the model you want to delete –> use view_results() for referece

Model_id : id of the model –> use view_results() for referece Flag : If true than will print out the contents of the db.

Example

mlog.delete_model(Model_name = "Mymodel") 
mlog.delete_model(Model_id = 1)

delete_all()

Flag : If true than will print out the contents of the db.

Example

mlog.delete_all()

model_profiles()

batch_size : How many entries to consider at once while comparing via graphs If batch_size less than total number of entries , than it will grouped into different pages . Different pages will have different graphs .If "All" is used than it will consider all the entries at once

port: use to change the port number in which the server is running.

debug: If true than run the server in debug mode

Note:

Do not run in debug mode while using it in jupyter notebook For jupyter notebook click on kernel/interrupt to stop the server .
for cmd/anaconda prompt ----> use [ctrl+c] to stop the server .

Example

mlog.model_profiles('All')
mlog.model_profiles(5)

Fun Part

All the graphs and insights are interlinked with the datatable . You can do the following operations out of the box and the graphs will change dynamically (as long as the server is running) :

Support model-logger

The development of model-logger relies completely on contributions.

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update tests as appropriate.

First roll out

May 31, 2020 💘

License

apache-2.0