Dev/zj (#10)

* Update Tutorial.md

* fix CPU train --> GPU test, GPU train --> CPU test. But CPU train --> multi-GPU test will report error.

* remove ujson

* remove glove for model_zoo/demo/conf.json

* modify glove for Tutorial.md

* add nltk.download('punkt') in problem.py

LearningMachine.py

@@ -667,12 +667,13 @@ class LearningMachine(object):
             self.model = torch.load(model_path)
             if isinstance(self.model, nn.DataParallel):
                 self.model = self.model.module
+            self.model.update_use_gpu(self.use_gpu)
             self.model = nn.DataParallel(self.model)
         else:
             self.model = torch.load(model_path, map_location='cpu')
             if isinstance(self.model, nn.DataParallel):
                 self.model = self.model.module
-            self.model.use_gpu = False
+            self.model.update_use_gpu(self.use_gpu)
 
         logging.info("Model %s loaded!" % model_path)
         logging.info("Total trainable parameters: %d" % (get_trainable_param_num(self.model)))

Model.py

@@ -6,7 +6,6 @@ import torch.nn as nn
 from block_zoo import *
 import copy
 import logging
-import ujson as json
 from utils.exceptions import ConfigurationError, LayerUndefinedError, LayerConfigUndefinedError
 from queue import Queue
 from utils.common_utils import transform_tensors2params, transfer_to_gpu
@@ -388,6 +387,18 @@ class Model(nn.Module):
     def is_cuda(self):
         return next(self.parameters()).data.is_cuda
 
+    def update_use_gpu(self, new_use_gpu):
+        self.use_gpu = new_use_gpu
+        for layer_id in self.layers.keys():
+            if isinstance(self.layers[layer_id], Embedding):
+                for input_cluster in self.layers[layer_id].embeddings:
+                    if isinstance(self.layers[layer_id].embeddings[input_cluster], CNNCharEmbedding):
+                        self.layers[layer_id].embeddings[input_cluster].layer_conf.use_gpu = new_use_gpu
+            elif isinstance(self.layers[layer_id], EncoderDecoder):
+                self.layers[layer_id].encoder.layer_conf.use_gpu = new_use_gpu
+                self.layers[layer_id].decoder.layer_conf.use_gpu = new_use_gpu
+            else:
+                self.layers[layer_id].layer_conf.use_gpu = new_use_gpu
 
 
 

ModelConf.py

@@ -2,7 +2,7 @@
 # Licensed under the MIT license.
 
 import codecs
-import ujson as json
+import json
 import os
 import tempfile
 import random

Tutorial.md

@@ -25,7 +25,7 @@
 
 ## <span id="installation">Installation</span>
 
-*Note: NeuronBlocks is based on Python 3.6*
+*Note: NeuronBlocks is based on **Python 3.6***
 
 1. Clone this project. 
     ```bash
@@ -37,7 +37,7 @@
     pip install -r requirements.txt
     ```
 
-3. Install PyTorch (*NeuronBlocks supports PyTorch version 0.4.1 currently*).
+3. Install PyTorch (*NeuronBlocks supports **PyTorch 0.4.1** currently*).
     
     For **Linux**, run the following command:
     ```bash
@@ -50,15 +50,11 @@
 
 ## <span id="quick-start">Quick Start</span>
 
-Get started by trying the given examples.
+Get started by trying the given examples. For **Windows**, we suggest you to use PowerShell instead of CMD.
 
 *Tips: in the following instruction, PROJECTROOT denotes the root directory of this project.*
 
 ```bash
-# get GloVe pre-trained word vectors
-cd PROJECT_ROOT/dataset
-bash get_glove.sh
-
 # train
 cd PROJECT_ROOT
 python train.py --conf_path=model_zoo/demo/conf.json
@@ -207,6 +203,11 @@ A model visualizer is provided for visualization and configuration correctness c
 
 In Model Zoo, we provide a suite of NLP models for common NLP tasks, in the form of JSON configuration files. You can pick one of existing models (JSON config files) in Model Zoo to start model training quickly, or build your own models by modifying the JSON config file to suit your specific task.
 
+*Note: Before trying models in NLP tasks, please download [GloVe](https://nlp.stanford.edu/projects/glove/) firstly via following commands.*
+```bash
+cd PROJECT_ROOT/dataset
+./get_glove.sh
+```
 
 ### <span id="task-1">Task 1: Text Classification</span>
 

block_zoo/BaseLayer.py

@@ -1,7 +1,6 @@
 # Copyright (c) Microsoft Corporation. All rights reserved.
 # Licensed under the MIT license.
 
-import ujson as json
 import codecs
 import logging
 

block_zoo/embedding/CNNCharEmbedding.py

@@ -99,6 +99,8 @@ class CNNCharEmbedding(BaseLayer):
         string_reshaped = string.view(string.size()[0], -1)     #[batch_size, seq_len * char num in words]
         char_embs_lookup = self.char_embeddings(string_reshaped).float()    # [batch_size, seq_len * char num in words, embedding_dim]
         if self.is_cuda():
+            if self.filters.device == torch.device('cpu'):
+                self.filters = self.filters.cuda()
             char_embs_lookup = char_embs_lookup.cuda(device=self.filters.device)
         char_embs_lookup = char_embs_lookup.view(-1, string.size()[2], self.layer_conf.embedding_matrix_dim)    #[batch_size * seq_len, char num in words, embedding_dim]
 

model_zoo/demo/conf.json

@@ -9,13 +9,9 @@
       "train_data_path": "./dataset/demo/train.tsv",
       "valid_data_path": "./dataset/demo/valid.tsv",
       "test_data_path": "./dataset/demo/test.tsv",
-      "predict_data_path": "./dataset/demo/predict.tsv",
-      "pre_trained_emb": "./dataset/GloVe/glove.840B.300d.txt"
+      "predict_data_path": "./dataset/demo/predict.tsv"
     },
     "file_with_col_header": false,
-    "pretrained_emb_type": "glove",
-    "pretrained_emb_binary_or_text": "text",
-    "involve_all_words_in_pretrained_emb": false,
     "add_start_end_for_seq": true,
     "file_header": {
       "question_text": 0,

problem.py

@@ -6,6 +6,8 @@ import numpy as np
 from core.CellDict import CellDict
 from tqdm import tqdm
 from utils.corpus_utils import load_embedding
+import nltk
+nltk.download('punkt')
 from utils.BPEEncoder import BPEEncoder
 import codecs
 from multiprocessing import cpu_count

requirements.txt

@@ -3,7 +3,6 @@ six==1.11.0
 nltk==3.4
 gensim==3.2.0
 tqdm==4.19.5
-ujson==1.35
 numpy==1.13.3
 scikit_learn==0.19.2
 PyYAML==3.13