Plumbing for CSV import from online logs

TODO.md

@@ -18,9 +18,8 @@
   - Use (anchor, positive) pairs to train a new model 
   - Use (sentence) + class labels to train a new model
 - Implement BatchAllTripletLoss
-- Implement a two-phase training regime 
-  1. Train with anchored definitions then...
-  2. Train with class labels
+- Train with anchored definitions and/or...
+- Train with class labels
 
 
 ## Evaluate

ucsinfer/__main__.py

@@ -1,11 +1,8 @@
 import os
-# import csv
 import logging
 from itertools import chain
 
-import tqdm
 import click
-# from tabulate import tabulate, SEPARATING_LINE
 
 from .inference import InferenceContext, load_ucs
 from .gather import (build_sentence_class_dataset, print_dataset_stats, 
@@ -136,6 +133,26 @@ def recommend(ctx, text, paths, interactive, skip_ucs):
                 os.rename(path, new_path)
                 break
         
+@ucsinfer.command('csv')
+@click.option('--filename-col', default="FileName", 
+              help="Heading or index of the column containing filenames",
+              show_default=True)
+@click.option('--description-col', default="TrackDescription", 
+              help="Heading or index of the column containing descriptions",
+              show_default=True)
+@click.option('--out', default='dataset/', show_default=True)
+@click.argument('paths', nargs=-1)
+@click.pass_context
+def csv(ctx, paths, out, filename_col, description_col):
+    """
+    Scan training data from CSV files 
+
+    `csv` is used to build a training dataset for finetuning the selected 
+    model, as like the `gather` command, except instead of scanning the 
+    file system it builds a dataset from descriptions and UCS filenames in 
+    columns of a CSV file.
+    """
+    pass
 
 @ucsinfer.command('gather')
 @click.option('--out', default='dataset/', show_default=True)
@@ -146,7 +163,7 @@ def recommend(ctx, text, paths, interactive, skip_ucs):
 @click.pass_context
 def gather(ctx, paths, out, ucs_data):
     """
-    Scan files to build a training dataset
+    Scan training data from audio files
     
     `gather` is used to build a training dataset for finetuning the selected
     model. Description sentences and UCS categories are collected from '.wav'