validation checks for memory nets and reading comprehension

examples/kvmemn2n/train_kvmemn2n.py

@@ -27,6 +27,8 @@
 
 from nlp_architect.models.kvmemn2n import KVMemN2N
 from nlp_architect.data.wikimovies import WIKIMOVIES
+from nlp_architect.utils.io import validate, validate_existing_directory, \
+ validate_existing_filepath, validate_parent_exists, check_size
 from interactive_util import interactive_loop
 
 from tqdm import tqdm
@@ -37,10 +39,13 @@
 # parse the command line arguments
 parser = NgraphArgparser(__doc__)
 parser.add_argument('--emb_size', type=int, default='50',
- help='Size of the word-embedding used in the model. (default 50)')
+ help='Size of the word-embedding used in the model. (default 50)',
+ action=check_size(1, 20000))
 parser.add_argument('--nhops', type=int, default='3',
- help='Number of memory hops in the network')
-parser.add_argument('--lr', type=float, default=0.01, help='learning rate')
+ help='Number of memory hops in the network',
+ action=check_size(1, 20))
+parser.add_argument('--lr', type=float, default=0.01, help='learning rate',
+ action=check_size(0, 5))
 parser.add_argument('--subset', type=str, default='wiki-entities',
  choices=['full', 'wiki-entities'],
  help='wikiMovies dataset to use for training examples.')
@@ -51,7 +56,8 @@
 parser.add_argument('--use_v_luts', action="store_true",
  help="Run the model using separate value lookup tables for each hop")
 parser.add_argument('--model_file', type=str, default=None,
- help="File to save or load weights from")
+ help="File to save or load weights from",
+ action=validate_parent_exists)
 parser.add_argument('--inference', action="store_true", help="Run Inference with loaded weight")
 parser.add_argument('--restore', action="store_true",
  help="Run the model restoring weights from model_file")

examples/kvmemn2n/wikiwindows.py

@@ -54,6 +54,9 @@
 from multiprocessing import Process, Queue, Condition, Value
 import os
 
+from nlp_architect.utils.io import validate, validate_existing_directory, \
+ validate_existing_filepath, validate_parent_exists, check_size
+
 '''
 Things that were changed from the original:
 1) Entities are replaced by their keys both in the key and the value (center word)
@@ -120,14 +123,18 @@
  description='Generates windowed examples for wikipedia files. By default,' +
  ' creates pairs of window<TAB>entity when used with entities.'
 )
-parser.add_argument('data_dir', type=str, help='name of root directory for files')
+parser.add_argument('data_dir', type=str, help='name of root directory for files',
+ action=validate_existing_directory)
 # parser.add_argument('input_file', type=str, nargs='+',
 # help='name of a input file in memnns format')
 parser.add_argument('-o', '--output_file', type=str,
- help='name of a output file, otherwise data_dir will be used')
-parser.add_argument('-n', type=int, help='Max number of examples to process.')
+ help='name of a output file, otherwise data_dir will be used',
+ action=validate_parent_exists)
+parser.add_argument('-n', type=int, help='Max number of examples to process.',
+ action=check_size(1, 100000000))
 parser.add_argument('-e', '--entities', type=str,
- help='entities file (each line specifies ngrams to always chunk together)')
+ help='entities file (each line specifies ngrams to always chunk together)',
+ action=validate_existing_filepath)
 parser.add_argument('-a', '--all_windows', action='store_true',
  help='if set, keeps all windows (not just ones entities). defaults to ' +
  ' True if entities file not present, False if it is present.')
@@ -150,7 +157,8 @@
  'sentence "hello world how are things" creates a window of "2:hello ' +
  '1:world <NULL> 1:are 2:things"')
 parser.add_argument('-t', '--num_threads', type=int, default=4,
- help='number of threads to use')
+ help='number of threads to use',
+ action=check_size(1,20))
 args = vars(parser.parse_args())
 
 beg = time.time()

examples/memn2n_dialogue/interactive.py

@@ -52,6 +52,8 @@
 from nlp_architect.models.memn2n_dialogue import MemN2N_Dialog
 from memn2n_dialogue.utils import interactive_loop
 from nlp_architect.utils.io import sanitize_path
+from nlp_architect.utils.io import validate, validate_existing_directory, \
+ validate_existing_filepath, validate_parent_exists, check_size
 
 # parse the command line arguments
 parser = NgraphArgparser(__doc__)
@@ -67,9 +69,11 @@
  '--emb_size',
  type=int,
  default='32',
- help='Size of the word-embedding used in the model. (default 128)')
+ help='Size of the word-embedding used in the model. (default 128)',
+ action=check_size(1, 2000))
 parser.add_argument('--nhops', type=int, default='3',
- help='Number of memory hops in the network')
+ help='Number of memory hops in the network',
+ action=check_size(1, 20))
 parser.add_argument(
  '--use_match_type',
  default=False,
@@ -94,11 +98,13 @@
  '--eps',
  type=float,
  default=1e-8,
- help='epsilon used to avoid divide by zero in softmax renormalization.')
+ help='epsilon used to avoid divide by zero in softmax renormalization.',
+ action=check_size(1e-100,1e-2))
 parser.add_argument(
  '--model_file',
  default='memn2n_weights.npz',
- help='File to load model weights from.')
+ help='File to load model weights from.',
+ action=validate_existing_filepath)
 
 parser.set_defaults(batch_size=32, epochs=200)
 args = parser.parse_args()

examples/memn2n_dialogue/train_model.py

@@ -53,7 +53,8 @@
 from tqdm import tqdm
 from utils import interactive_loop
 from nlp_architect.utils.io import sanitize_path
-
+from nlp_architect.utils.io import validate, validate_existing_directory, \
+ validate_existing_filepath, validate_parent_exists, check_size
 
 # parse the command line arguments
 parser = NgraphArgparser(__doc__)
@@ -69,9 +70,11 @@
  '--emb_size',
  type=int,
  default='32',
- help='Size of the word-embedding used in the model. (default 128)')
+ help='Size of the word-embedding used in the model. (default 32)',
+ action=check_size(1, 2000))
 parser.add_argument('--nhops', type=int, default='3',
- help='Number of memory hops in the network')
+ help='Number of memory hops in the network',
+ action=check_size(1, 20))
 parser.add_argument(
  '--use_match_type',
  default=False,
@@ -97,12 +100,14 @@
  '--grad_clip_norm',
  type=float,
  default=40.0,
- help='Clip gradients such that norm is below this value.')
+ help='Clip gradients such that norm is below this value.',
+ action=check_size(0,100))
 parser.add_argument(
  '--eps',
  type=float,
  default=1e-8,
- help='epsilon used to avoid divide by zero in softmax renormalization.')
+ help='epsilon used to avoid divide by zero in softmax renormalization.',
+ action=check_size(1e-100,1e-2))
 parser.add_argument(
  '--save_log',
  action='store_true',
@@ -112,17 +117,20 @@
  '--log_file',
  type=str,
  default='memn2n_dialgoue_results.txt',
- help='File to write evaluation set results to.')
+ help='File to write evaluation set results to.',
+ action=validate_parent_exists)
 parser.add_argument(
  '--weights_save_path',
  type=str,
  default='memn2n_weights.npz',
- help='File to save model weights to.')
+ help='File to save model weights to.',
+ action=validate_parent_exists)
 parser.add_argument(
  '--save_epochs',
  type=int,
  default=1,
- help='Number of epochs between saving model weights.')
+ help='Number of epochs between saving model weights.',
+ action=check_size(1,1000))
 parser.add_argument(
  '--restore',
  default=False,

examples/reading_comprehension/prepare_data.py

@@ -22,6 +22,8 @@
 import os
 from tqdm import *
 from nlp_architect.utils.io import sanitize_path
+from nlp_architect.utils.io import validate, validate_existing_directory, \
+ validate_existing_filepath, validate_parent_exists, check_size
 
 PAD = "<pad>"
 SOS = "<sos>"
@@ -199,13 +201,15 @@ def get_ids_list(data_lists, vocab):
  parser.add_argument(
  '--data_path',
  default='',
- help='enter path where training data and the glove embeddings were downloaded')
+ help='enter path where training data and the glove embeddings were downloaded',
+ action=validate_existing_directory)
 
  parser.add_argument(
  '--preprocess_glove',
  type=int,
  default=1,
- help='Chose whether or not to preprocess glove embeddings')
+ help='Chose whether or not to preprocess glove embeddings',
+ action=check_size(0,1))
 
  parser.set_defaults()
 

examples/reading_comprehension/train.py

@@ -42,9 +42,10 @@
  get_data_array_squad_ngraph,
  cal_f1_score)
 import math
-from nlp_architect.utils.weight_initilizers import (make_placeholder, make_weights)
+from nlp_architect.contrib.ngraph.weight_initilizers import (make_placeholder, make_weights)
 from nlp_architect.utils.io import sanitize_path
-
+from nlp_architect.utils.io import validate, validate_existing_directory, \
+ validate_existing_filepath, validate_parent_exists, check_size
 
 """
 Model converges on gpu backened.
@@ -55,18 +56,22 @@
 
 parser.add_argument(
  '--data_path',
- help='enter path for training data')
+ help='enter path for training data',
+ action=validate_existing_directory)
 
 parser.add_argument('--gpu_id', default="0",
- help='enter gpu id')
+ help='enter gpu id',
+ action=check_size(0,10))
 
 parser.add_argument('--max_para_req', default=100,
- help='enter the max length of paragraph')
+ help='enter the max length of paragraph',
+ action=check_size(0,10000))
 
 parser.add_argument(
  '--batch_size_squad',
  default=16,
- help='enter the batch size') # 16 is the max batch size o be fit on gpu
+ help='enter the batch size',
+ action=check_size(1,10000)) # 16 is the max batch size o be fit on gpu
 
 parser.set_defaults()
 

nlp_architect/models/kvmemn2n.py

@@ -21,7 +21,7 @@
 from ngraph.frontends.neon import Layer
 from ngraph.frontends.neon import GaussianInit
 from ngraph.frontends.neon.graph import SubGraph
-from nlp_architect.utils.encodings import position_encoding
+from nlp_architect.contrib.ngraph.encodings import position_encoding
 from nlp_architect.contrib.ngraph.modified_lookup_table import ModifiedLookupTable
 
 

examples/memn2n_dialogue/tests.py → tests/test_memn2n_dialogue.py

@@ -27,29 +27,13 @@
 from ngraph.frontends.neon import ax
 import pytest
 
-pytestmark = pytest.mark.transformer_dependent
-
-
 rng = RandomTensorGenerator(0, np.float32)
 
 delta = 1e-8
 rtol = atol = 1e-2
 
-
-def pytest_generate_tests(metafunc):
- bsz_rng = [1]
-
- if 'lut_args' in metafunc.fixturenames:
- fargs = []
- vocab_rng = [10, 50, 100]
- embed_rng = [20, 18, 31]
- # Add third axis (memory size) to test multiaxis LUT
- mem_rng = [1, 5, 19]
- bsz_rng = [4, 32]
- seq_rng = [3, 5]
- fargs = itt.product(vocab_rng, embed_rng, bsz_rng, seq_rng, mem_rng)
- metafunc.parametrize('lut_args', fargs)
-
+(V, F, N, T, M) = (3, 6, 4, 5, 7)
+lut_args = (V, F, N, T, M)
 
 def lut_fprop_ref(lut, idx):
  """
@@ -71,7 +55,9 @@ def lut_update_ref(error, lut, idx, pad_idx):
  dw_ref[wrd_id, :] = np.sum(error.take(group[0], axis=0), axis=0)
  return dw_ref
 
-
+@pytest.mark.parametrize('show_tok', [True, False])
+@pytest.mark.parametrize('show_doc', [True, False])
+@pytest.mark.parametrize('lut_args', [lut_args])
 def test_lut(lut_args):
  """
  test lut fprop and bprop
@@ -127,11 +113,3 @@ def test_lut(lut_args):
  pad_idx=pad_idx)
  ng.testing.assert_allclose(
  update_lut, update_ref, rtol=0.0, atol=1.0e-5)
-
-
-if __name__ == '__main__':
- factory = ngt.make_transformer_factory('cpu')
- ngt.set_transformer_factory(factory)
- (V, F, N, T, M) = (3, 6, 4, 5, 7)
- lut_args = (V, F, N, T, M)
- test_lut(lut_args)

tests/test_np_semantic_segmentation.py

@@ -11,7 +11,7 @@ def test_model_training():
  Test model end2end training
  """
  data_path = os.path.abspath('fixtures/data/np_semantic_segmentation_prepared_data.csv')
- model_path = os.path.abspath('fixtures/data/np_semantic_segmentation')
+ model_path = os.path.abspath('fixtures/data/np_semantic_segmentation.prm')
  num_epochs = 200
  be = gen_backend(batch_size=64)
  # load data sets from file
@@ -45,4 +45,3 @@ def test_model_inference():
  assert len(output_reader_list) == len(input_reader_list) - 1
  os.remove(model_path)
  os.remove(output_path)
-