add tensorflow-gpu to conda env, fix ch10_translate and unify futil.r…

…ead_text read_txt

conda/environment.yml

@@ -33,9 +33,9 @@ dependencies:
   - scipy==1.1.0
   - spacy==2.0.18
   - swig>=3.0.12
-  - tensorflow>=1.13.1
+  - tensorflow-gpu>=1.13.1
   - tensorflow-hub>=0.4.0
-  - theano==1.0.2
+  - theano>=1.0.2
   # this will fail on windows:
   # - python-annoy>=1.9.5,<2.0.0
 

src/nlpia/book/examples/ch10_translate.py

@@ -102,6 +102,10 @@
 from keras.callbacks import ModelCheckpoint  # noqa
 from nlpia.constants import BIGDATA_PATH  # noqa
 checkpoint_path = os.path.join(BIGDATA_PATH, 'checkpoints')
+try:
+    os.mkdir(checkpoint_path)
+except FileExistsError:
+    pass
 checkpoint_path = os.path.join(checkpoint_path, 'nlpia-seq2seq-translation-weights.{epoch:02d}-{val_loss:.2f}.hdf5')
 
 

src/nlpia/futil.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
-""" File utilities comparable to similarly named bash utils: rm_rf(), rm_f(), and mkdir_p() 
+""" File utilities comparable to similarly named bash utils: rm_rf(), rm_f(), and mkdir_p()
 
 dataset1.0 is in files like: PPE1.rar PPE2.zip PPE3.zip PP4.7zip
 dataset2.0 is in gs:/Buckets/safety_monitoring/data/obj/supplemental/"""
@@ -123,7 +123,7 @@ def rm_r(path, force=False):
     elif os.path.isdir(path):
         try:
             return os.rmdir(path)
-        except OSError:  # OSError: [Errno 66] Directory not empty: 
+        except OSError:  # OSError: [Errno 66] Directory not empty:
             pass
         except:
             if not force:
@@ -223,7 +223,7 @@ def ensure_open(f, mode='r'):
                 return open(f, mode=mode)
             f = fin  # reset path in case it is the text that needs to be opened with StringIO
         else:
-            f = io.StringIO(f) 
+            f = io.StringIO(f)
     elif f and getattr(f, 'closed', None):
         if hasattr(f, '_write_gzip_header'):
             return gzip.open(f.name, mode=mode)
@@ -341,11 +341,11 @@ def update_dict_types(d, update_keys=True, update_values=True, typ=(int,)):
         di[ki] = vi
     d.update(di)
     return d
-            
+
 
 def read_json(filepath, intkeys=True, intvalues=True):
     """ read text from filepath (`open(find_filepath(expand_filepath(fp)))`) then json.loads()
-    
+
     >>> read_json('HTTP_1.1  Status Code Definitions.html.json')
     {'100': 'Continue',
      '101': 'Switching Protocols',...
@@ -432,6 +432,3 @@ def read_text(forfn, nrows=None, verbose=True):
                 ) / float(len(lines)) > .05:
             return np.array(html2text(EOL.join(lines)).split(EOL))
     return lines
-
-
-read_txt = read_text

src/nlpia/loaders.py

@@ -312,7 +312,7 @@ def load_anki_df(language='deu'):
         'http://www.cs.cornell.edu/~cristian/data/cornell_movie_dialogs_corpus.zip',
         9916637,
         'cornell_movie_dialogs_corpus',
-        
+
     ),
     'save_dialog_tweets': (
         'https://www.dropbox.com/s/tlrr9bm45uzm9yl/save_dialog_tweets.txt.gz?dl=1',
@@ -981,7 +981,7 @@ def read_named_csv(name, data_path=DATA_PATH, nrows=None, verbose=True):
         except (IOError, pd.errors.ParserError):
             pass
         try:
-            return read_txt(name, nrows=nrows)
+            return read_text(name, nrows=nrows)
         except (IOError, UnicodeDecodeError):
             pass
     data_path = expand_filepath(data_path)
@@ -1006,7 +1006,7 @@ def read_named_csv(name, data_path=DATA_PATH, nrows=None, verbose=True):
     except IOError:
         pass
     try:
-        return read_txt(os.path.join(data_path, name + '.txt'), verbose=verbose)
+        return read_text(os.path.join(data_path, name + '.txt'), verbose=verbose)
     except IOError:
         pass
 
@@ -1019,7 +1019,7 @@ def read_named_csv(name, data_path=DATA_PATH, nrows=None, verbose=True):
     except ValueError:
         pass
     try:
-        return read_txt(os.path.join(BIGDATA_PATH, name + '.txt'), verbose=verbose)
+        return read_text(os.path.join(BIGDATA_PATH, name + '.txt'), verbose=verbose)
     except IOError:
         pass
 
@@ -1093,7 +1093,7 @@ def get_data(name='sms-spam', nrows=None, limit=None):
                 pass
         if filepathlow.endswith('.txt'):
             try:
-                return read_txt(filepath)
+                return read_text(filepath)
             except (TypeError, UnicodeError):
                 pass
         return filepaths[name]
@@ -1290,7 +1290,7 @@ def cleaner(row):
 
 def clean_cornell_movies(filename='cornell_movie_dialogs_corpus.zip', subdir='cornell movie-dialogs corpus'):
     """ Load a dataframe of ~100k raw (uncollated) movie lines from the cornell movies dialog corpus
-    
+
     >>> local_filepath = download_file(BIG_URLS['cornell_movie_dialogs_corpus'][0])
     >>> df = clean_cornell_movies(filename='cornell_movie_dialogs_corpus.zip')
     >>> df.describe(include='all')