Upload 16 files

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "roberta-large",
+  "_name_or_path": "img_intents_model",
   "architectures": [
     "RobertaForSequenceClassification"
   ],
@@ -21,7 +21,7 @@
   "position_embedding_type": "absolute",
   "problem_type": "single_label_classification",
   "torch_dtype": "float32",
-  "transformers_version": "4.31.0",
+  "transformers_version": "4.32.0.dev0",
   "type_vocab_size": 1,
   "use_cache": true,
   "vocab_size": 50265

confirmTrain.py

@@ -0,0 +1,78 @@
+from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments
+from datasets import Dataset, load_from_disk, concatenate_datasets
+import os
+import torch
+import numpy as np
+from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, confusion_matrix
+
+MODEL_NAME = "roberta-large"
+SAVE_MODEL_FOLDER = "img_intents_model"
+OUTPUT_DIR = "./results"
+NEG_NAME = "NEGATIVE"
+POS_NAME = "POSITIVE"
+
+# Load the model and tokenizer
+model = AutoModelForSequenceClassification.from_pretrained(SAVE_MODEL_FOLDER)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+
+# Load the training arguments
+training_args = torch.load(os.path.join(OUTPUT_DIR, "training_args.bin"))
+
+# Load the sentences from the text files into lists
+with open('test_positives.txt', 'r') as file:
+    positives_texts = [line.strip() for line in file.readlines()]
+with open('test_negatives.txt', 'r') as file:
+    negatives_texts = [line.strip() for line in file.readlines()]
+
+# Create datasets from the lists and add a 'label' column
+positives_dataset = Dataset.from_dict({'text': positives_texts, 'label': [1]*len(positives_texts)})
+negatives_dataset = Dataset.from_dict({'text': negatives_texts, 'label': [0]*len(negatives_texts)})
+
+# Combine into a single dataset
+test_dataset = concatenate_datasets([positives_dataset, negatives_dataset])
+
+# Preprocessing function
+def preprocess_function(examples):
+    # Tokenize the texts
+    return tokenizer(examples["text"], truncation=True, max_length=512, padding='max_length')
+
+test_dataset = test_dataset.map(preprocess_function, batched=True)
+
+# Make sure all your tensors are the same size for batching together
+test_dataset = test_dataset.remove_columns(["text"]).rename_column("label", "labels").with_format("torch")
+
+# Create the Trainer object
+trainer = Trainer(
+    model=model,
+    args=training_args,
+)
+
+# Evaluate the model and save predictions and labels
+predictions, labels, _ = trainer.predict(test_dataset)
+
+# Convert predictions to binary (0 or 1)
+binary_predictions = np.argmax(predictions, axis=1)
+
+# Print overall metrics
+accuracy = accuracy_score(labels, binary_predictions)
+precision = precision_score(labels, binary_predictions)
+recall = recall_score(labels, binary_predictions)
+f1 = f1_score(labels, binary_predictions)
+
+print(f"Overall accuracy: {accuracy}")
+print(f"Overall precision: {precision}")
+print(f"Overall recall: {recall}")
+print(f"Overall F1 score: {f1}")
+
+# Print the report for each class
+cm = confusion_matrix(labels, binary_predictions)
+
+for i, class_name in enumerate([NEG_NAME, POS_NAME]):
+    total = cm[i].sum()
+    correct = cm[i][i]
+    loss = total - correct
+
+    print(f"\n{class_name}:")
+    print(f"Total: {total}")
+    print(f"Confirmed: {correct}")
+    print(f"Loss: {loss} ({loss / total * 100:.2f}%)")

events.out.tfevents.1690445575.DESKTOP-HMQHN82.32493.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b6c685b93f5d55fd18eabefb6dccf600c9d603bd2fecbb5dab6845e115284ff
+size 4775

fineTune.py

@@ -0,0 +1,68 @@
+from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments
+from datasets import load_dataset, concatenate_datasets
+
+MODEL_NAME = "roberta-large"
+SAVE_MODEL_FOLDER = "img_intents_model"
+OUTPUT_DIR = "./results"
+output_dir = "/results"
+
+# Load the model and tokenizer
+model = AutoModelForSequenceClassification.from_pretrained(SAVE_MODEL_FOLDER)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+
+# Load the sentences from the text files into datasets
+positives_dataset = load_dataset('text', data_files='test_positives.txt')
+negatives_dataset = load_dataset('text', data_files='test_negatives.txt')
+
+# Manually assign split names to the datasets
+positives_dataset = positives_dataset['train'].map(lambda example: {'label': 1})
+negatives_dataset = negatives_dataset['train'].map(lambda example: {'label': 0})
+
+# Combine into a single dataset and add a 'label' column
+train_dataset = concatenate_datasets([positives_dataset, negatives_dataset])
+
+# Preprocessing function
+def preprocess_function(examples):
+    # Tokenize the texts
+    return tokenizer(examples["text"], truncation=True, max_length=512, padding='max_length')
+
+train_dataset = train_dataset.map(preprocess_function, batched=True)
+
+# Make sure all your tensors are the same size for batching together
+train_dataset = train_dataset.remove_columns(["text"]).rename_column("label", "labels").with_format("torch")
+
+# TrainingArguments
+training_args = TrainingArguments(
+    output_dir=OUTPUT_DIR,
+    num_train_epochs=5,  # Fine-tune for a few epochs
+    per_device_train_batch_size=16,  # Decrease this if necessary
+    per_device_eval_batch_size=64,
+    warmup_steps=500,
+    weight_decay=0.01,
+    logging_dir=OUTPUT_DIR,
+    logging_strategy='steps',  # Log after every training step
+    logging_steps=10,  # Adjust this to change how often logging occurs
+    evaluation_strategy='steps',  # Evaluate after every training step
+    eval_steps=100,  # Adjust this to change how often evaluation occurs
+    save_strategy='steps',  # Save after every training step
+    save_steps=500,  # Adjust this to change how often saving occurs
+    no_cuda=False,  # Use GPU
+    gradient_accumulation_steps=2,  # If necessary
+    fp16=True,  # Use mixed precision training
+    report_to='tensorboard'
+)
+# Create a Trainer
+trainer = Trainer(
+    model=model,
+    args=training_args,
+    train_dataset=train_dataset,
+)
+
+# Fine-tune the model
+trainer.train()
+
+# Save the model
+trainer.save_model(SAVE_MODEL_FOLDER)
+
+# Save the tokenizer
+tokenizer.save_pretrained(OUTPUT_DIR)

img_intents.py

@@ -0,0 +1,60 @@
+import random
+from datasets import Dataset, DatasetDict
+import pandas as pd
+
+# Load positive examples from 'Positives.txt'
+with open('Positives.txt', 'r') as file:
+    positive_examples = [line.strip() for line in file.readlines()]
+
+# Load negative examples from 'Negatives.txt'
+with open('Negatives.txt', 'r') as file:
+    negative_examples = [line.strip() for line in file.readlines()]
+
+# Shuffle and combine positive and negative examples
+all_examples = [(example, 'POSITIVE') for example in positive_examples] + [(example, 'NEGATIVE') for example in negative_examples]
+random.shuffle(all_examples)
+
+# Convert to pandas DataFrame
+df = pd.DataFrame(all_examples, columns=['text', 'label'])
+
+# Split the dataset if desired (e.g., 80% train, 10% validation, 10% test)
+train_size = int(0.8 * len(df))
+val_size = int(0.1 * len(df))
+train_examples = df[:train_size]
+val_examples = df[train_size: train_size + val_size]
+test_examples = df[train_size + val_size:]
+
+# Save the dataset to CSV format with 'split' column
+import csv
+
+train_examples['split'] = 'train'
+val_examples['split'] = 'validation'
+test_examples['split'] = 'test'
+
+with open('dataset_with_split.csv', 'w', newline='', encoding='utf-8') as csvfile:
+    csvwriter = csv.writer(csvfile)
+    csvwriter.writerow(['text', 'label', 'split'])  # Write header
+    csvwriter.writerows(train_examples.values.tolist())  # Write train examples
+    csvwriter.writerows(val_examples.values.tolist())  # Write validation examples
+    csvwriter.writerows(test_examples.values.tolist())  # Write test examples
+
+print("Dataset with 'split' column created successfully.")
+
+# Load the dataset from the CSV file
+dataset = Dataset.from_csv('dataset_with_split.csv')
+
+# Create a DatasetDict object containing train, validation, and test datasets
+datasets = DatasetDict({
+    'train': dataset.filter(lambda example: example['split'] == 'train'),
+    'validation': dataset.filter(lambda example: example['split'] == 'val'),
+    'test': dataset.filter(lambda example: example['split'] == 'test'),
+})
+
+# Optional: Define dataset metadata
+dataset_info = {
+    "name": "img_intents",
+    "description": "A dataset of positive and negative examples",
+    "citation": "Provide the citation or source of the dataset",
+    "homepage": "Link to the dataset homepage",
+}
+

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1ff2702abff612a3c3aadfbfaae3ecc5e54b0203d92f0c94f7e3b38b559cbf0
+oid sha256:bf5d6bcff602ec06c6251c09acf42c6a6e3039a782060be4accd9a5a0c723d42
 size 1421582769

test.py

@@ -0,0 +1,32 @@
+import torch
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+
+MODEL_DIR = "img_intents_model"
+TOKENIZER_NAME = "./results"
+
+# Load the trained model
+model = AutoModelForSequenceClassification.from_pretrained(MODEL_DIR)
+
+# Load the tokenizer
+tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_NAME)
+
+while True:
+    # Get the input from the command line
+    input_text = input("Enter a message to classify (or 'q' to quit): ")
+
+    if input_text.lower() == 'q':
+        break
+
+    # Encode the input and convert it to a torch tensor
+    inputs = tokenizer.encode_plus(input_text, return_tensors='pt')
+
+    # Get the model's prediction
+    outputs = model(**inputs)
+
+    # Get the predicted class from the model's output
+    predicted_class = torch.argmax(outputs.logits).item()
+
+    if predicted_class == 1:
+        print("The message is predicted as an image intent.")
+    else:
+        print("The message is not predicted as an image intent.")

test_negatives.txt

@@ -0,0 +1,65 @@
+Hey, how's it going?
+U up for a movie tonight?
+What's the weather like tmrw?
+Can't make it to the party, sorry.
+Did u finish the assignment?
+Hey, just wanted to check on you. How have you been?
+Did you hear about the latest update on the app?
+I hope everything is going well on your end.
+Could we possibly reschedule our meeting for tomorrow?
+Just a quick reminder about our lunch date.
+Hey, did you manage to catch the latest episode?
+I've been thinking about taking a vacation. Any recommendations?
+How's the new job going? Must be exciting!
+Remember our trip last summer? Good times!
+Been a while since we caught up. Coffee soon?
+Good day. I hope this message finds you well.
+I regret to inform you of a delay in the project.
+Thank you for your prompt response to our previous correspondence.
+Could you kindly provide an update on the matter?
+I appreciate your assistance in this matter.
+Did you know that honey never spoils?
+I've recently taken up reading as a hobby. It's quite relaxing.
+The concert last night was absolutely phenomenal.
+I've always been fascinated by the stars and constellations.
+Nature walks have a therapeutic effect, don't you think?
+I've been feeling a bit under the weather lately.
+Did you hear about the new restaurant downtown?
+Trust the process, everything will fall into place.
+Books are a uniquely portable magic.
+The world is full of endless possibilities.
+How's your family doing?
+I've been thinking of adopting a pet. What do you think?
+Music is such a powerful form of expression.
+There's a meteor shower happening next week.
+Did you hear about the recent developments in the tech world?
+Life has a way of surprising us when we least expect it.
+The movie last night was a roller coaster of emotions.
+Have you ever tried meditation?
+I've been meaning to learn a new language. It's never too late, right?
+Nature is truly the best artist.
+R u going to the concert tonight?
+What's the homework for tomorrow?
+Did u see the new episode last night?
+Pizza for dinner?
+U seen my keys?
+Could we possibly move the meeting to Friday?
+Did you manage to get the tickets for the game?
+Hope everything is well with you.
+Don't forget about our trip next week!
+Just a reminder about the appointment tomorrow.
+Hey, did you watch the football match yesterday?
+Been thinking about starting a new hobby. Any ideas?
+How's the new course you signed up for?
+Remember the high school days? So nostalgic!
+Been a while since our last road trip. We should plan one soon.
+Dear Sir, I hope this message finds you in good health.
+Regrettably, there will be a delay in the delivery.
+Thank you for your prompt attention to this matter.
+Could you kindly provide an update on the project?
+I appreciate your patience in this matter.
+Ever tried bungee jumping? It's quite an adrenaline rush.
+I've recently started a book club. You should join!
+The new art exhibition downtown is quite impressive.
+I've always found the ocean to be quite calming.
+There's something magical about a good cup of coffee, don't you think?

test_positives.txt

@@ -0,0 +1,60 @@
+Hey, got any pics of dogs? Need one for a project.
+Can u find a pic of the Grand Canyon? Wanna use it as a wallpaper.
+I need a graphic of the solar system for my presentation. Can u help?
+Hey, can u get me a pic of a beach sunset? Need it for a collage.
+Show me a picture of a Ferrari, wanna show my friends.
+Could you find an image of a beautiful waterfall for me? Thanks!
+I'm looking for a photo of a bustling cityscape. Could you assist?
+Could you please locate a picture of the Sydney Opera House for me?
+I need a graphic of the human skeleton for my study, could you help?
+I'm searching for a photo of a coral reef. Can you assist?
+Hey, I'm looking for a cool pic of a sports car. Can you find one?
+I'm in need of a picture of a blue whale. Can you help me out?
+Hey, could you find a photo of a rainbow for me? Would really appreciate it.
+I'm after a graphic of a world map. Can you help?
+Can you find a photo of a tropical island for me? Thanks in advance!
+Could you please provide an image of the Eiffel Tower?
+I require a picture of a beautiful garden for my project.
+Could you assist in locating a photograph of the Great Wall of China?
+I'm in need of a graphic of the human brain for my research. Could you assist?
+Could you please locate a picture of the Northern Lights for me?
+I need an image of a serene lake. Can you assist?
+Could you provide a graphic of the universe?
+I'm looking for a pic of a wild animal. Can you help?
+Can you find a photo of a bustling cityscape for me?
+I need a picture of an elephant. Can you assist?
+Can you fetch me a photo of a delicious meal?
+Could you help me locate a picture of a medieval castle?
+I'm searching for a photo of a scenic landscape. Can you assist?
+Can you find a graphic of a human heart for me?
+I'm after a picture of a tranquil forest. Can you help me out?
+Yo, got a pic of a golden retriever? Need it for a meme.
+U got a photo of the Pyramids of Giza? Need it for a report.
+Need a graphic of the Milky Way for my project. Can u help?
+Hey, can u find me a pic of a snowy mountain? Need it for a mood board.
+Show me a picture of a Lamborghini, wanna make my friends jealous.
+Could you find an image of cherry blossoms for me? Thanks!
+I'm looking for a photo of a lively marketplace. Could you assist?
+Could you please locate a picture of the Colosseum for me?
+I need a graphic of the human body for my study, could you help?
+I'm searching for a photo of a desert. Can you assist?
+Hey, I'm looking for a cool pic of a yacht. Can you find one?
+I'm in need of a picture of a grizzly bear. Can you help me out?
+Hey, could you find a photo of a lightning strike for me? Would really appreciate it.
+I'm after a graphic of the solar system. Can you help?
+Can you find a photo of a bustling street for me? Thanks in advance!
+Could you please provide an image of the Leaning Tower of Pisa?
+I require a picture of a serene river for my project.
+Could you assist in locating a photograph of the Amazon rainforest?
+I'm in need of a graphic of the human digestive system for my research. Could you assist?
+Could you please locate a picture of a glacier for me?
+I need an image of a bustling airport. Can you assist?
+Could you provide a graphic of the Earth's crust?
+I'm looking for a pic of a majestic lion. Can you help?
+Can you find a photo of a crowded concert for me?
+I need a picture of a giraffe. Can you assist?
+Can you fetch me a photo of a scrumptious pizza?
+Could you help me locate a picture of a haunted house?
+I'm searching for a photo of a rolling hillside. Can you assist?
+Can you find a graphic of the human nervous system for me?
+I'm after a picture of a tranquil beach. Can you help me out?

tokenizer.json

@@ -1,21 +1,7 @@
 {
   "version": "1.0",
-  "truncation": {
-    "direction": "Right",
-    "max_length": 512,
-    "strategy": "LongestFirst",
-    "stride": 0
-  },
-  "padding": {
-    "strategy": {
-      "Fixed": 512
-    },
-    "direction": "Right",
-    "pad_to_multiple_of": null,
-    "pad_id": 1,
-    "pad_type_id": 0,
-    "pad_token": "<pad>"
-  },
+  "truncation": null,
+  "padding": null,
   "added_tokens": [
     {
       "id": 0,

train.py

@@ -0,0 +1,139 @@
+from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments
+from datasets import Dataset, DatasetDict, load_from_disk, concatenate_datasets
+import torch
+import os
+import tensorflow as tf
+
+OUTPUT_DIR = "./results"
+DATASET_NAME = 'dataset_with_split.csv'
+MODEL_NAME = "roberta-large"
+LOG_DIR = "./logs"
+SAVE_MODEL_FOLDER = "img_intents_model"
+
+POS_NAME = "POSITIVE"
+NEG_NAME = "NEGATIVE"
+
+
+BATCH_SIZE_TRAIN = 16
+BATCH_SIZE_EVAL = 64
+EPOCS = 10
+WARMUP_STEPS = 500
+
+
+tf.debugging.experimental.enable_dump_debug_info(
+    LOG_DIR,
+    tensor_debug_mode="FULL_HEALTH",
+    circular_buffer_size=1000,
+    op_regex=None,
+    tensor_dtypes=None
+)
+
+# Load the dataset from the CSV file
+dataset = Dataset.from_csv(DATASET_NAME)
+
+# Create a DatasetDict object containing train, validation, and test datasets
+datasets = DatasetDict({
+    'train': dataset.filter(lambda example: example['split'] == 'train'),
+    'validation': dataset.filter(lambda example: example['split'] == 'validation'),
+    'test': dataset.filter(lambda example: example['split'] == 'test'),
+})
+
+# Balance the datasets
+for split in datasets.keys():
+    num_positive = len(datasets[split].filter(lambda example: example['label'] == POS_NAME))
+    num_negative = len(datasets[split].filter(lambda example: example['label'] == NEG_NAME))
+
+    if num_positive > num_negative:
+        # Downsample the positive examples
+        datasets[split] = concatenate_datasets([
+            datasets[split].filter(lambda example: example['label'] == POS_NAME).shuffle(seed=42).select(range(num_negative)),
+            datasets[split].filter(lambda example: example['label'] == NEG_NAME)
+        ])
+    else:
+        # Downsample the negative examples
+        datasets[split] = concatenate_datasets([
+            datasets[split].filter(lambda example: example['label'] == POS_NAME),
+            datasets[split].filter(lambda example: example['label'] == NEG_NAME).shuffle(seed=42).select(range(num_positive))
+        ])
+
+    # Shuffle the dataset to mix positive and negative examples
+    datasets[split] = datasets[split].shuffle(seed=42)
+
+# Specify the model name
+model_name = MODEL_NAME  # Or whatever model you want to use
+
+# Load the tokenizer associated with your model
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+# Load your datasets
+train_dataset = datasets['train']
+val_dataset = datasets['validation']
+test_dataset = datasets['test']
+
+# Preprocessing function
+def preprocess_function(examples):
+    # Replace None in 'text' field with an empty string
+    examples["text"] = [text if text is not None else "" for text in examples["text"]]
+
+    # Convert labels from string to int
+    examples["label"] = [1 if label == POS_NAME else 0 for label in examples["label"]]
+
+    # Tokenize the texts
+    return tokenizer(examples["text"], truncation=True, max_length=512, padding='max_length')
+
+train_dataset = train_dataset.map(preprocess_function, batched=True)
+val_dataset = val_dataset.map(preprocess_function, batched=True)
+test_dataset = test_dataset.map(preprocess_function, batched=True)
+
+# Make sure all your tensors are the same size for batching together
+train_dataset = train_dataset.remove_columns(["text"]).rename_column("label", "labels").with_format("torch")
+val_dataset = val_dataset.remove_columns(["text"]).rename_column("label", "labels").with_format("torch")
+test_dataset = test_dataset.remove_columns(["text"]).rename_column("label", "labels").with_format("torch")
+
+# Load a pre-trained model for sequence classification
+model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2)  # You have two labels: POSITIVE and NEGATIVE
+
+# TrainingArguments
+training_args = TrainingArguments(
+    output_dir=OUTPUT_DIR,
+    num_train_epochs=EPOCS,
+    per_device_train_batch_size=BATCH_SIZE_TRAIN,  # decrease this if necessary
+    per_device_eval_batch_size=BATCH_SIZE_EVAL,
+    warmup_steps=WARMUP_STEPS,
+    weight_decay=0.01,
+    logging_dir=LOG_DIR,
+    logging_strategy='steps',  # Log after every training step
+    logging_steps=10,  # Adjust this to change how often logging occurs
+    evaluation_strategy='steps',  # Evaluate after every training step
+    eval_steps=100,  # Adjust this to change how often evaluation occurs
+    save_strategy='steps',  # Save after every training step
+    save_steps=500,  # Adjust this to change how often saving occurs
+    no_cuda=False,  # use GPU
+    gradient_accumulation_steps=2,  # if necessary
+    fp16=True,  # use mixed precision training
+    report_to='tensorboard'
+)
+
+# Create a Trainer
+trainer = Trainer(
+    model=model,
+    args=training_args,
+    train_dataset=train_dataset,
+    eval_dataset=val_dataset,
+)
+
+# Train the model
+trainer.train()
+
+# Save the model
+trainer.save_model(SAVE_MODEL_FOLDER)
+
+# Save the tokenizer
+tokenizer.save_pretrained(OUTPUT_DIR)
+
+# Save the training arguments
+torch.save(training_args, os.path.join(OUTPUT_DIR, "training_args.bin"))
+
+# Evaluate the model and print the results
+eval_results = trainer.evaluate(test_dataset)
+print(f"Test set evaluation results: {eval_results}")

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99cb2d4f988e976319f712a4a50a4eb0e7485053e6ce66f46cd8987070543c66
+oid sha256:8434cb556cc1edbd48d4ac1a3f3ec0b1e09492c025297d4c4ef93d53462acec8
 size 3899