import streamlit as st import pandas as pd from sklearn.model_selection import train_test_split from sklearn.ensemble import RandomForestRegressor from sklearn.metrics import mean_squared_error from mp_api.client import MPRester import time from matminer.featurizers.conversions import StrToComposition from matminer.featurizers.composition import ElementProperty import plotly.express as px import plotly.graph_objects as go # 请替换为你自己的 Material Project API 密钥 API_KEY = "bLi1JNyQysy3R4OdR9DZYj1beCkK1mru" # 从 Material Project 爬取数据 def fetch_materials_data(elements, num_data): try: with MPRester(API_KEY) as mpr: progress_bar = st.progress(0) status_text = st.empty() if num_data == "全部": all_docs = [] chunk_size = 1000 offset = 0 while True: docs = mpr.materials.summary.search(elements=elements, num_chunks=chunk_size, offset=offset) if not docs: break all_docs.extend(docs) offset += chunk_size else: num_data = int(num_data) all_docs = mpr.materials.summary.search(elements=elements, num_chunks=num_data) # 手动筛选数据 if num_data != "全部": all_docs = all_docs[:num_data] total_docs = len(all_docs) data = [] for i, doc in enumerate(all_docs): entry = { "formula": doc.formula_pretty, "energy_per_atom": doc.energy_per_atom, "formation_energy_per_atom": doc.formation_energy_per_atom, "band_gap": doc.band_gap } data.append(entry) progress = (i + 1) / total_docs progress_bar.progress(progress) status_text.text(f"正在搜索材料数据: {int(progress * 100)}%") status_text.text("材料数据搜索完成！") df = pd.DataFrame(data) return df except Exception as e: st.error(f"在从 Material Project 获取数据时发生错误: {e}") return pd.DataFrame() # 处理 formula 转换为特征值 def convert_formula_to_features(data): if 'formula' in data.columns: st.write("正在将 formula 转换为特征值...") # 将字符串化学式转换为 Composition 对象 stc = StrToComposition() data = stc.featurize_dataframe(data, "formula") # 添加元素属性特征 ep_feat = ElementProperty.from_preset(preset_name="magpie") data = ep_feat.featurize_dataframe(data, col_id="composition") st.write("formula 转换为特征值完成。") return data # 模拟自动机器学习功能 def auto_ml_function(data, predict_variable): if data.empty: st.warning("未获取到有效数据，请检查输入的元素。") return None, None, None try: # 筛选出数值列 numeric_columns = data.select_dtypes(include=['number']).columns if predict_variable not in numeric_columns: st.error(f"需要预测值 {predict_variable} 不是数值类型，请重新选择。") return None, None, None X = data[numeric_columns.drop(predict_variable)] y = data[predict_variable] # 划分训练集和测试集 X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42) # 机器学习进度条 progress_bar = st.progress(0) status_text = st.empty() model = RandomForestRegressor() # 模拟训练进度 for i in range(100): time.sleep(0.01) progress_bar.progress(i + 1) status_text.text(f"模型训练中: {i + 1}%") status_text.text("模型训练完成！") progress_bar.empty() model.fit(X_train, y_train) # 预测 y_pred = model.predict(X_test) mse = mean_squared_error(y_test, y_pred) rmse = mse ** 0.5 return rmse, y_test, y_pred, model.feature_importances_, X.columns except Exception as e: st.error(f"模型训练过程中出现错误: {e}") return None, None, None def main(): # 初始化会话状态 if 'step' not in st.session_state: st.session_state.step = 1 if 'elements' not in st.session_state: st.session_state.elements = [] if 'data' not in st.session_state: st.session_state.data = pd.DataFrame() if 'predict_variable' not in st.session_state: st.session_state.predict_variable = None if 'num_data' not in st.session_state: st.session_state.num_data = "100" st.title("自动机器学习演示") st.markdown("此应用允许你输入材料元素名字，从 Material Project 爬取数据，运行自动机器学习模型，并查看模型的准确率。") if st.session_state.step == 1: # 用户输入材料元素名字 elements_input = st.text_input("输入想要计算的材料元素名字，用逗号分隔（例如：Fe,O）", ','.join(st.session_state.elements)) st.session_state.elements = [element.strip() for element in elements_input.split(",") if element.strip()] # 用户选择提取的数据量 st.session_state.num_data = st.selectbox("选择提取的数据量", ["100", "200", "500", "全部"]) if st.session_state.elements: if st.button("从 Material Project 爬取数据"): st.session_state.step = 2 if st.session_state.step == 2: with st.spinner("正在从 Material Project 爬取数据..."): st.session_state.data = fetch_materials_data(st.session_state.elements, st.session_state.num_data) if not st.session_state.data.empty: # 转换 formula 为特征值 st.session_state.data = convert_formula_to_features(st.session_state.data) st.write("数据基本信息：") st.session_state.data.info() st.write("数据前几行信息：") st.dataframe(st.session_state.data.head()) # 选择需要预测值 numeric_columns = st.session_state.data.select_dtypes(include=['number']).columns st.session_state.predict_variable = st.selectbox("选择需要预测值", numeric_columns) if st.session_state.predict_variable: if st.button("运行自动机器学习"): st.session_state.step = 3 if st.session_state.step == 3: rmse, y_test, y_pred, feature_importances, feature_names = auto_ml_function(st.session_state.data, st.session_state.predict_variable) if rmse is not None: st.write("自动机器学习模型均方根误差：") st.metric(label="均方根误差", value=f"{rmse:.2f}") # 绘制预测值与真实值的散点图 fig1 = px.scatter(x=y_test, y=y_pred, labels={'x': '真实值', 'y': '预测值'}, title='预测值与真实值的散点图') fig1.add_trace(go.Scatter(x=[y_test.min(), y_test.max()], y=[y_test.min(), y_test.max()], mode='lines', name='完美预测线')) st.plotly_chart(fig1) # 绘制特征重要性柱状图 feature_importance_df = pd.DataFrame({ '特征': feature_names, '重要性': feature_importances }) feature_importance_df = feature_importance_df.sort_values(by='重要性', ascending=False) fig2 = px.bar(feature_importance_df, x='特征', y='重要性', title='特征重要性柱状图') st.plotly_chart(fig2) st.session_state.step = 1 # 完成后回到第一步 if __name__ == "__main__": main()