Under construction.
modelTrainer.py builds a machine-learning model using a logistic regression classifier, if given a dataset with the filename companies_sorted.csv.
In our dataset of about 7,000,000 companies, the following table shows the approximate number of companies per each size range:
| size_range | Number of companies |
|---|---|
| 1 - 10 | 1674694 |
| 11 - 50 | 570370 |
| 51 - 200 | 181524 |
| 201 - 500 | 43749 |
| 501 - 1000 | 15647 |
| 1001 - 5000 | 13632 |
| 5001 - 10000 | 1855 |
| 10001+ | 1366 |
The dataset contains around 7 million companies, with attributes like company names, company locations, year of inception, size category, etc.000
modelTrainer.py creates a machine-learning model based on this data. The model, given a company name and year of inception, tries its best to predict that company's size category (see above).
companySizeEstimator.py is the frontend for this model.
python modelTrainer.py [--testonly] [filename]
The [--testonly] flag specifies to test an existing machine-learning model, instead of creating and training one. If using the --testonly flag, you have to specify a [filename] from which to load the existing model.
I wrote this from scratch, with the exception of some borrowed snippets from the Intro to Artifical Intelligence course at uOttawa (CSI 4506, Prof. Caroline Barrière) - specifically the train_model() method came from a homework assignment notebook.