From 862cedb4878b8f45aa18e269bd07a121a1eff3fb Mon Sep 17 00:00:00 2001
From: Ivan Pavlovich <ivan.pavlovic@hes-so.ch>
Date: Fri, 21 Mar 2025 10:52:30 +0100
Subject: [PATCH] =?UTF-8?q?Ajout=20de=20m=C3=A9triques=20suppl=C3=A9mentai?=
=?UTF-8?q?re=20dans=20le=20calcule=20de=20teste=20et=20ajout=20de=20la=20?=
=?UTF-8?q?LLM=20host=C3=A9e=20Cohere?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
---
models/LLM/Cohere/cohereCommand.py | 20 ---------
models/LLM/Cohere/cohere_wrapper.py | 36 +++++++++++++++
testModel/metrics.py | 44 +++++++++++++++++--
...aurer-DeBERTa_v3_base_mnli_fever_anli.json | 0
...-deberta_v3_base_zeroshot_v1_1_all_33.json | 0
...er-multilingual_MiniLMv2_L6_mnli_xnli.json | 0
testModel/results/{ => v1}/deepseek_v2.json | 0
.../{ => v1}/facebook-bart_large_mnli.json | 0
testModel/results/{ => v1}/gemini_hosted.json | 0
testModel/results/{ => v1}/llama3_2.json | 0
testModel/results/{ => v1}/mistral_small.json | 0
testModel/test.py | 24 ++++------
testModel/utils.py | 40 +++++++++++------
variables/models.py | 4 +-
14 files changed, 113 insertions(+), 55 deletions(-)
delete mode 100644 models/LLM/Cohere/cohereCommand.py
create mode 100644 models/LLM/Cohere/cohere_wrapper.py
rename testModel/results/{ => v1}/MoritzLaurer-DeBERTa_v3_base_mnli_fever_anli.json (100%)
rename testModel/results/{ => v1}/MoritzLaurer-deberta_v3_base_zeroshot_v1_1_all_33.json (100%)
rename testModel/results/{ => v1}/MoritzLaurer-multilingual_MiniLMv2_L6_mnli_xnli.json (100%)
rename testModel/results/{ => v1}/deepseek_v2.json (100%)
rename testModel/results/{ => v1}/facebook-bart_large_mnli.json (100%)
rename testModel/results/{ => v1}/gemini_hosted.json (100%)
rename testModel/results/{ => v1}/llama3_2.json (100%)
rename testModel/results/{ => v1}/mistral_small.json (100%)
diff --git a/models/LLM/Cohere/cohereCommand.py b/models/LLM/Cohere/cohereCommand.py
deleted file mode 100644
index 86e4fe160..000000000
--- a/models/LLM/Cohere/cohereCommand.py
+++ /dev/null
@@ -1,20 +0,0 @@
-import cohere
-import json
-
-def cohere_create_client():
- return cohere.ClientV2("hNdIrXFRGTzxSK4bnEXONMMIYukcMuPqOLHV6P28")
-
-def cohere_classify(client, sequence, labels):
- prompt = f'I need you to give me the labels that could be given to the text (keep in mind that u can put multiple labels and select only the labels that i give you):\
- text: {sequence}\
- labels: {labels}\
- Give the response in json format {{ "labels": [] }}'
-
- response = client.chat(
- model="command-r-plus",
- messages=[{"role": "user", "content": prompt}]
- )
-
- responce_json = json.loads(response.message.content[0].text)
-
- return responce_json
\ No newline at end of file
diff --git a/models/LLM/Cohere/cohere_wrapper.py b/models/LLM/Cohere/cohere_wrapper.py
new file mode 100644
index 000000000..d32892a1e
--- /dev/null
+++ b/models/LLM/Cohere/cohere_wrapper.py
@@ -0,0 +1,36 @@
+import cohere
+import json
+import time
+
+def cohere_create_client():
+ return cohere.ClientV2("hNdIrXFRGTzxSK4bnEXONMMIYukcMuPqOLHV6P28")
+
+def cohere_classify(client, sequence, labels):
+ prompt = 'I need you to give me the labels that could be given to the text (keep in mind that u can put multiple labels and select only the labels that i give you):\n'
+ prompt += 'text: ' + sequence + '\n'
+ prompt += 'labels: ' + "[" + ", ".join(labels) + ']\n'
+ prompt += 'Give the response in json format: { "labels": [] }. Follow this writing to the letter and don t add text around it. Only put the labels that you select between [].'
+ prompt += 'Even if you have no labels give me a response like: { "labels": [] }. And also put labels between "".'
+
+ response = client.chat(
+ model="command-r-plus",
+ messages=[{"role": "user", "content": prompt}]
+ )
+
+ json_str = response.message.content[0].text.strip().lstrip('```json').strip()
+
+ if json_str.endswith("```"):
+ json_str = json_str[0:-3]
+
+ json_str = json_str.split("}")[0] + "}"
+
+ responce_json = json.loads(json_str)
+
+ res = {}
+
+ for label in labels:
+ res[label] = label in responce_json["labels"]
+
+ time.sleep(6)
+
+ return res
\ No newline at end of file
diff --git a/testModel/metrics.py b/testModel/metrics.py
index d9b79d0ef..f12da370e 100644
--- a/testModel/metrics.py
+++ b/testModel/metrics.py
@@ -1,19 +1,49 @@
-def confusion_matrix(wanted, prediction):
+def get_confusion_matrix(wanted, prediction):
matrix = [[0, 0], [0, 0]]
for key in wanted.keys():
- if wanted[key]:
- if prediction[key]:
+
+ actual = wanted[key]
+ predicted = prediction[key]
+
+ if key in ["Diabetes type 1", "Diabetes type 2"]:
+ if predicted:
+ if actual is False:
+ if wanted["Diabetes"]:
+ actual = True
+ elif key == "Diabetes":
+ if predicted:
+ if actual is False:
+ if wanted["Diabetes type 1"] or wanted["Diabetes type 2"]:
+ actual = True
+
+ if actual:
+ if predicted:
matrix[0][0] += 1
else:
matrix[1][0] += 1
else:
- if prediction[key]:
+ if predicted:
matrix[0][1] += 1
else:
matrix[1][1] += 1
return matrix
+def get_label_confusion_matrix(label, wanted, prediction):
+ matrix = [[0, 0], [0, 0]]
+
+ if wanted[label]:
+ if prediction[label]:
+ matrix[0][0] += 1
+ else:
+ matrix[1][0] += 1
+ else:
+ if prediction[label]:
+ matrix[0][1] += 1
+ else:
+ matrix[1][1] += 1
+ return matrix
+
def add_confusion_matrices(confusion_matrix, tmp_confusion_matrix):
for i in range(2):
for j in range(2):
@@ -39,6 +69,12 @@ def get_precision(confusion_matrix):
return 0
return confusion_matrix[0][0] / denominator
+def get_error_rate(confusion_matrix):
+ denominator = (confusion_matrix[0][0] + confusion_matrix[0][1] + confusion_matrix[1][0] + confusion_matrix[1][1])
+ if denominator == 0:
+ return 0
+ return (confusion_matrix[0][1] + confusion_matrix[1][0]) / denominator
+
def get_f1_score(confusion_matrix):
precision = get_precision(confusion_matrix)
recall = get_tpr(confusion_matrix)
diff --git a/testModel/results/MoritzLaurer-DeBERTa_v3_base_mnli_fever_anli.json b/testModel/results/v1/MoritzLaurer-DeBERTa_v3_base_mnli_fever_anli.json
similarity index 100%
rename from testModel/results/MoritzLaurer-DeBERTa_v3_base_mnli_fever_anli.json
rename to testModel/results/v1/MoritzLaurer-DeBERTa_v3_base_mnli_fever_anli.json
diff --git a/testModel/results/MoritzLaurer-deberta_v3_base_zeroshot_v1_1_all_33.json b/testModel/results/v1/MoritzLaurer-deberta_v3_base_zeroshot_v1_1_all_33.json
similarity index 100%
rename from testModel/results/MoritzLaurer-deberta_v3_base_zeroshot_v1_1_all_33.json
rename to testModel/results/v1/MoritzLaurer-deberta_v3_base_zeroshot_v1_1_all_33.json
diff --git a/testModel/results/MoritzLaurer-multilingual_MiniLMv2_L6_mnli_xnli.json b/testModel/results/v1/MoritzLaurer-multilingual_MiniLMv2_L6_mnli_xnli.json
similarity index 100%
rename from testModel/results/MoritzLaurer-multilingual_MiniLMv2_L6_mnli_xnli.json
rename to testModel/results/v1/MoritzLaurer-multilingual_MiniLMv2_L6_mnli_xnli.json
diff --git a/testModel/results/deepseek_v2.json b/testModel/results/v1/deepseek_v2.json
similarity index 100%
rename from testModel/results/deepseek_v2.json
rename to testModel/results/v1/deepseek_v2.json
diff --git a/testModel/results/facebook-bart_large_mnli.json b/testModel/results/v1/facebook-bart_large_mnli.json
similarity index 100%
rename from testModel/results/facebook-bart_large_mnli.json
rename to testModel/results/v1/facebook-bart_large_mnli.json
diff --git a/testModel/results/gemini_hosted.json b/testModel/results/v1/gemini_hosted.json
similarity index 100%
rename from testModel/results/gemini_hosted.json
rename to testModel/results/v1/gemini_hosted.json
diff --git a/testModel/results/llama3_2.json b/testModel/results/v1/llama3_2.json
similarity index 100%
rename from testModel/results/llama3_2.json
rename to testModel/results/v1/llama3_2.json
diff --git a/testModel/results/mistral_small.json b/testModel/results/v1/mistral_small.json
similarity index 100%
rename from testModel/results/mistral_small.json
rename to testModel/results/v1/mistral_small.json
diff --git a/testModel/test.py b/testModel/test.py
index 4e4626890..0ace50201 100644
--- a/testModel/test.py
+++ b/testModel/test.py
@@ -13,7 +13,7 @@ from variables.huggingface import HUGGINGFACE_MODELS
from variables.articles import LENGTH_CATEGORIES, LENGTH_CATEGORIES_TRESHOLDS
from variables.models import MODELS
from testModel.utils import get_dataset_filename, get_article_data, get_wanted_predictions, save_model_json, save_model_results_json, create_model_data_var, store_results
-from testModel.metrics import confusion_matrix, add_confusion_matrices, get_tpr, get_tnr, get_precision
+from testModel.metrics import get_confusion_matrix
from parsers.jsonParser import parseJsonFile
from models.ZeroShotClassifier.HuggingFace.zero_shot_classification import create_classifier, classify
@@ -53,7 +53,7 @@ else:
print(f"File {file_path} does not exist.")
if not os.path.exists(file_path):
- initial_data = create_model_data_var()
+ initial_data = create_model_data_var(NCDS)
print(initial_data)
with open(file_path, "w", encoding="utf-8") as file:
json.dump(initial_data, file, indent=4)
@@ -77,12 +77,6 @@ print(f"MODEL: {model}")
print(f"TRESHOLD: {data['treshold']}")
print("---------------------------------")
-result_matrix = data["results"]["ALL"]["confusion matrix"]
-length_matrix = {}
-
-for length_category in LENGTH_CATEGORIES:
- length_matrix[length_category] = data["results"][length_category]["confusion matrix"]
-
for ncd in NCDS:
try:
@@ -118,7 +112,6 @@ for ncd in NCDS:
end = time.time()
selected_labels = [key for key, value in predictions.items() if value]
- matrix = confusion_matrix(wanted, predictions)
article_result["PMID"] = article['PMID']
article_result["Labels"] = article["Predictions"]
@@ -126,7 +119,7 @@ for ncd in NCDS:
article_result["Wanted"] = wanted
article_result["Predictions"] = predictions
article_result["Selected Labels"] = selected_labels
- article_result["confusion matrix"] = matrix
+ article_result["confusion matrix"] = get_confusion_matrix(wanted, predictions)
data["articles"].append(article_result)
print(f"PMID: {article_result['PMID']}")
@@ -137,20 +130,17 @@ for ncd in NCDS:
print(f"Selected labels: {article_result['Selected Labels']}")
print(f"Confusion matrix: {article_result['confusion matrix']}")
- result_matrix = add_confusion_matrices(result_matrix, matrix)
- data["results"]["ALL"] = store_results(result_matrix)
+ data["results"]["ALL"] = store_results(data["results"]["ALL"], wanted, predictions)
added = False
for id, length_category_treshold in enumerate(LENGTH_CATEGORIES_TRESHOLDS):
if len(title+abstract) < length_category_treshold:
- length_matrix[LENGTH_CATEGORIES[id]] = add_confusion_matrices(length_matrix[LENGTH_CATEGORIES[id]], matrix)
- data["results"][LENGTH_CATEGORIES[id]] = store_results(length_matrix[LENGTH_CATEGORIES[id]])
+ data["results"][LENGTH_CATEGORIES[id]] = store_results(data["results"][LENGTH_CATEGORIES[id]], wanted, predictions)
added = True
break
if not added:
- length_matrix[LENGTH_CATEGORIES[-1]] = add_confusion_matrices(length_matrix[LENGTH_CATEGORIES[-1]], matrix)
- data["results"][LENGTH_CATEGORIES[-1]] = store_results(length_matrix[LENGTH_CATEGORIES[-1]])
+ data["results"][LENGTH_CATEGORIES[-1]] = store_results(data["results"][LENGTH_CATEGORIES[id]], wanted, predictions)
print("---------------------------------")
@@ -166,6 +156,7 @@ for length_category in LENGTH_CATEGORIES:
print(f"True Positive Rate (TPR): {data['results'][length_category]['TPR']}")
print(f"True Negative Rate (TNR): {data['results'][length_category]['TNR']}")
print(f"Precision: {data['results'][length_category]['Precision']}")
+ print(f"Error Rate: {data['results'][length_category]['Error rate']}")
print(f"F1 Score: {data['results'][length_category]['F1']}")
print("---------------------------------")
@@ -175,5 +166,6 @@ print(f"Result confusion matrix: {data['results']['ALL']['confusion matrix']}")
print(f"True Positive Rate (TPR): {data['results']['ALL']['TPR']}")
print(f"True Negative Rate (TNR): {data['results']['ALL']['TNR']}")
print(f"Precision: {data['results']['ALL']['Precision']}")
+print(f"Error Rate: {data['results']['ALL']['Error rate']}")
print(f"F1 Score: {data['results']['ALL']['F1']}")
print()
\ No newline at end of file
diff --git a/testModel/utils.py b/testModel/utils.py
index 18b1dbec9..b54f2e51e 100644
--- a/testModel/utils.py
+++ b/testModel/utils.py
@@ -4,7 +4,7 @@ import json
sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), "../")))
-from testModel.metrics import get_tpr, get_tnr, get_precision, get_f1_score
+from testModel.metrics import get_confusion_matrix, get_label_confusion_matrix, get_tpr, get_tnr, get_precision, get_error_rate, get_f1_score, add_confusion_matrices
from variables.articles import LENGTH_CATEGORIES
def get_dataset_filename(name):
@@ -75,7 +75,7 @@ def save_model_results_json(filepath, model, new_entry):
with open(filename, "w") as file:
json.dump(data, file)
-def create_model_data_var():
+def create_model_data_var(labels):
data = {
'treshold': None,
'articles': [],
@@ -84,27 +84,39 @@ def create_model_data_var():
}
}
- data["results"]["ALL"] = create_results()
+ data["results"]["ALL"] = create_results(labels)
for length_category in LENGTH_CATEGORIES:
- data["results"][length_category] = create_results()
+ data["results"][length_category] = create_results(labels)
return data
-def create_results():
- return {
+def create_results(labels):
+ res = {
'confusion matrix': [[0, 0], [0, 0]],
+ 'labels': {},
'TPR': 0,
'TNR': 0,
'Precision': 0,
+ 'Error rate': 0,
'F1': 0
}
-def store_results(confusion_matrix):
- return {
- "confusion matrix": confusion_matrix,
- "TPR": get_tpr(confusion_matrix),
- "TNR": get_tnr(confusion_matrix),
- "Precision": get_precision(confusion_matrix),
- "F1": get_f1_score(confusion_matrix)
- }
\ No newline at end of file
+ for label in labels:
+ res["labels"][label] = [[0, 0], [0, 0]]
+
+ return res
+
+def store_results(results, wanted, predicted):
+ results["confusion matrix"] = add_confusion_matrices(results["confusion matrix"], get_confusion_matrix(wanted, predicted))
+
+ for label, confusion_matrix in results["labels"].items():
+ confusion_matrix = add_confusion_matrices(confusion_matrix, get_label_confusion_matrix(label, wanted, predicted))
+
+ results["TPR"] = get_tpr(results["confusion matrix"])
+ results["TNR"] = get_tnr(results["confusion matrix"])
+ results["Precision"] = get_precision(results["confusion matrix"])
+ results["Error rate"] = get_error_rate(results["confusion matrix"])
+ results["F1"] = get_f1_score(results["confusion matrix"])
+
+ return results
diff --git a/variables/models.py b/variables/models.py
index ec03a0300..f406410d1 100644
--- a/variables/models.py
+++ b/variables/models.py
@@ -7,6 +7,7 @@ sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '../')))
from models.ZeroShotClassifier.HuggingFace.zero_shot_classification import create_classifier, classify
import models.LLM.Ollama.ollama_wrapper as ollama
from models.LLM.Gemini.gemini import gemini_start_chat, gemini_classify
+from models.LLM.Cohere.cohere_wrapper import cohere_create_client, cohere_classify
MODELS_CATEGORIES = [
"huggingface",
@@ -23,5 +24,6 @@ MODELS = {
'llama3.2': {'category': "ollama", 'predict': ollama.classify, 'requiresDependency': False, 'dependency': None},
'mistral-small': {'category': "ollama", 'predict': ollama.classify, 'requiresDependency': False, 'dependency': None},
'deepseek-v2': {'category': "ollama", 'predict': ollama.classify, 'requiresDependency': False, 'dependency': None},
- 'gemini-hosted': {'category': "hosted", 'predict': gemini_classify, 'requiresDependency': True, 'dependency': gemini_start_chat}
+ 'gemini-hosted': {'category': "hosted", 'predict': gemini_classify, 'requiresDependency': True, 'dependency': gemini_start_chat},
+ 'cohere-hosted': {'category': "hosted", 'predict': cohere_classify, 'requiresDependency': True, 'dependency': cohere_create_client}
}
\ No newline at end of file
--
GitLab