diff --git a/ai.sp b/ai.sp
new file mode 100644
index 0000000..f4495d2
--- /dev/null
+++ b/ai.sp
@@ -0,0 +1,339 @@
+#pragma semicolon 1
+#pragma newdecls required
+
+/* SM Includes */
+#include <sourcemod>
+#include <sdktools>
+#include <sdktools_gamerules>
+#define TIME_TO_TICK(%1)    (RoundToNearest((%1) / GetTickInterval()))
+#define TICK_TO_TIME(%1)    ((%1) * GetTickInterval())
+#define IS_CLIENT(%1)       (1 <= %1 <= MaxClients)
+#define ABS(%1) ((%1)>0 ? (%1) : -(%1)) 
+/* Plugin Info */
+public Plugin myinfo =
+{
+    name = "Crow Aimbot Detector",
+    author = "huoji",
+    description = "Analyzes clients to detect aimbots",
+    version = "0.0.1",
+    url = "http://key08.com/"
+};
+
+/* Globals */
+#define AIM_ANGLE_CHANGE	15	// Max angle change that a player should snap
+#define AIM_BAN_MIN			4		// Minimum number of detections before an auto-ban is allowed
+#define AIM_MIN_DISTANCE	50.0	// Minimum distance acceptable for a detection.
+Handle g_IgnoreWeapons = INVALID_HANDLE;
+
+float g_fEyeAngles[MAXPLAYERS+1][64][3];
+
+float g_Sensitivity[MAXPLAYERS + 1];
+float g_mYaw[MAXPLAYERS + 1];
+float g_mPitch[MAXPLAYERS + 1];
+
+int g_iEyeIndex[MAXPLAYERS+1];
+int g_iMouse[MAXPLAYERS+1][64][2];
+int g_iAimDetections[MAXPLAYERS+1];
+int g_iAimbotBan = 0;
+int g_iMaxAngleHistory;
+int g_cheat = 0;
+bool g_bIsShoting[MAXPLAYERS+1][64];
+/* Plugin Functions */
+public void OnPluginStart()
+{
+    // Store no more than 300ms worth of angle history.
+    if ((g_iMaxAngleHistory = TIME_TO_TICK(0.3)) > sizeof(g_fEyeAngles[]))
+    {
+        g_iMaxAngleHistory = sizeof(g_fEyeAngles[]);
+    }
+
+    // Weapons to ignore when analyzing.
+    g_IgnoreWeapons = CreateTrie();
+
+    SetTrieValue(g_IgnoreWeapons, "weapon_knife", 1);
+    SetTrieValue(g_IgnoreWeapons, "weapon_taser", 1);
+
+    // Hooks.
+    HookEntityOutput("trigger_teleport", "OnEndTouch", Teleport_OnEndTouch);
+    HookEvent("player_spawn", Event_PlayerSpawn, EventHookMode_Post);
+    HookEvent("player_death", Event_PlayerDeath, EventHookMode_Post);
+    HookEventEx("cs_intermission", Event_Intermission);
+    g_cheat = CreateConVar("sm_crowai_cheat", "0", "0 = no cheat , 1 = cheating.", _, true, 0.0, true, 1.0);
+    PrintToServer("[CrowAi] Init Success!");
+}
+void OnFileUploadCallBack(HTTPStatus status, any value)
+{
+    if (status != HTTPStatus_OK) {
+        PrintToServer("文件上传失败");
+        return;
+    }
+
+    PrintToServer("Upload complete");
+} 
+public Action Event_Intermission(Handle event, const char[] name, bool dontBroadcast)
+{
+    PrintToChatAll("比赛结束!上传反作弊数据中!");
+    //上传数据
+	return Plugin_Continue;
+}
+stock bool IsClientNew(int client)
+{
+    // Client must be ingame.
+    return IsFakeClient(client) || GetGameTime() > GetClientTime(client);
+}
+
+stock void ZeroVector(float vec[3])
+{
+    vec[0] = vec[1] = vec[2] = 0.0;
+}
+
+stock bool IsVectorZero(const float vec[3])
+{
+    return vec[0] == 0.0 && vec[1] == 0.0 && vec[2] == 0.0;
+}
+
+public void OnClientPutInServer(int client)
+{
+    if (IsClientNew(client))
+    {
+        g_iAimDetections[client] = 0;
+        Aimbot_ClearAngles(client);
+    }
+}
+
+
+public void Teleport_OnEndTouch(const char[] output, int caller, int activator, float delay)
+{
+    /* A client is being teleported in the map. */
+    if (IS_CLIENT(activator) && IsClientConnected(activator))
+    {
+        Aimbot_ClearAngles(activator);
+        CreateTimer(0.1 + delay, Timer_ClearAngles, GetClientUserId(activator), TIMER_FLAG_NO_MAPCHANGE);
+    }
+}
+
+public Action Event_PlayerSpawn(Event event, const char[] name, bool dontBroadcast)
+{
+    int userid = event.GetInt("userid");
+    int client = GetClientOfUserId(userid);
+    
+    if (IS_CLIENT(client))
+    {
+        Aimbot_ClearAngles(client);
+        CreateTimer(0.1, Timer_ClearAngles, userid, TIMER_FLAG_NO_MAPCHANGE);
+    }
+}
+
+public Action Event_PlayerDeath(Event event, const char[] name, bool dontBroadcast)
+{
+    char sWeapon[32], dummy;
+    //GetEventString(event, "weapon", sWeapon, sizeof(sWeapon));
+    event.GetString("weapon", sWeapon, sizeof(sWeapon));
+
+    if (GetTrieValue(g_IgnoreWeapons, sWeapon, dummy))
+    {
+        return;
+    }
+    if (GameRules_GetProp("m_bWarmupPeriod") == 1)
+    {
+         return;
+    }
+    int victim = GetClientOfUserId(event.GetInt("userid"));
+    int attacker = GetClientOfUserId(event.GetInt("attacker"));
+    if (IS_CLIENT(attacker) && victim != attacker && IsClientInGame(attacker))
+    {
+        float vVictim[3], vAttacker[3];
+        GetClientAbsOrigin(victim, vVictim);
+        GetClientAbsOrigin(attacker, vAttacker);
+        if (GetVectorDistance(vVictim, vAttacker) >= AIM_MIN_DISTANCE)
+        {
+            Aimbot_AnalyzeAngles(attacker);
+        }
+    }
+}
+
+public Action Timer_ClearAngles(Handle timer, any userid)
+{
+    /* Delayed because the client's angles can sometimes "spin" after being teleported. */
+    int client = GetClientOfUserId(userid);
+    
+    if (IS_CLIENT(client))
+    {
+        Aimbot_ClearAngles(client);
+    }
+    
+    return Plugin_Stop;
+}
+
+public Action Timer_DecreaseCount(Handle timer, any userid)
+{
+    /* Decrease the detection count by 1. */
+    int client = GetClientOfUserId(userid);
+    
+    if (IS_CLIENT(client) && g_iAimDetections[client])
+    {
+        g_iAimDetections[client]--;
+    }
+    
+    return Plugin_Stop;
+}
+int m_iShotsFired(int client)
+{
+    return GetEntPropEnt(client, Prop_Send, "m_iShotsFired");
+}
+float m_aimPunchAngleX(int client)
+{
+    float punch[3];
+    GetEntPropVector(client, Prop_Send, "m_aimPunchAngle",punch);
+    return punch[0];
+}
+float m_aimPunchAngleY(int client)
+{
+    float punch[3];
+    GetEntPropVector(client, Prop_Send, "m_aimPunchAngle",punch);
+    return punch[1];
+}
+public Action OnPlayerRunCmd(int client, int& buttons, int& impulse, float vel[3], float angles[3], int& weapon, int& subtype, int& cmdnum, int& tickcount, int& seed, int mouse[2])
+{
+    g_fEyeAngles[client][g_iEyeIndex[client]] = angles;
+    g_iMouse[client][g_iEyeIndex[client]] = mouse;
+    g_bIsShoting[client][g_iEyeIndex[client]] = false;
+    if(m_iShotsFired(client) > 2)
+    {
+        g_fEyeAngles[client][g_iEyeIndex[client]][0] += m_aimPunchAngleX(client) * 2;
+        g_fEyeAngles[client][g_iEyeIndex[client]][1] += m_aimPunchAngleY(client) * 2;
+        g_bIsShoting[client][g_iEyeIndex[client]] = true;
+    }
+    
+    if (++g_iEyeIndex[client] == g_iMaxAngleHistory)
+    {
+        g_iEyeIndex[client] = 0;
+    }
+    return Plugin_Continue;
+}
+public void ConVar_QueryClient(QueryCookie cookie, int client, ConVarQueryResult result, const char[] cvarName, const char[] cvarValue)
+{
+    if(!IsFakeClient(client))
+	{
+		if(result == ConVarQuery_Okay)
+		{
+			if(StrEqual("sensitivity", cvarName))
+			{
+				g_Sensitivity[client] = StringToFloat(cvarValue);
+			}else if(StrEqual("m_yaw", cvarName))
+			{
+				g_mYaw[client] = StringToFloat(cvarValue);
+			}
+            else if(StrEqual("m_pitch", cvarName))
+			{
+				g_mPitch[client] = StringToFloat(cvarValue);
+			}
+        }
+    }
+}
+
+void Aimbot_AnalyzeAngles(int client)
+{
+    if(IsFakeClient(client))
+        return;
+    /* Analyze the client to see if their angles snapped. */
+    float vLastAngles[3], vAngles[3], fAngleDiff,Last_AngleDiff;
+    int idx = g_iEyeIndex[client];
+    QueryClientConVar(client, "sensitivity", ConVar_QueryClient, client);
+	QueryClientConVar(client, "m_yaw", ConVar_QueryClient, client);
+    QueryClientConVar(client, "m_pitch", ConVar_QueryClient, client);
+    int num_of_detect = 0;
+    int num_of_detect_smooth = 0;
+    //PrintToChatAll("[CrowAntiCheat] 开始分析 %N",client);
+    char steamid[64];
+    GetClientAuthId(client, AuthId_SteamID64, steamid, sizeof(steamid));
+    char path_name[PLATFORM_MAX_PATH];
+    Format(path_name, sizeof(path_name), "addons\\sourcemod\\ai\\ai_%s.csv",steamid);
+    bool file_exists = true;
+    if(!FileExists(path_name)){
+        file_exists = false;
+    }
+    Handle handle_file = OpenFile(path_name, "a+");
+    char file_head[19999];
+    if(file_exists == false){
+        bool first_meme = false;
+        for (int i = 0; i < g_iMaxAngleHistory; i++)
+        {
+            if (i == 0)
+            {
+                continue;
+            }
+            if(first_meme == false){
+                first_meme = true;
+                Format(file_head, sizeof(file_head), "%d_x,%d_y,%d_diff,%d_mx,%d_my",i,i,i,i,i);
+            }else{
+                Format(file_head, sizeof(file_head), "%s,%d_x,%d_y,%d_diff,%d_mx,%d_my",file_head,i,i,i,i,i);
+            }
+           
+        }
+        WriteFileLine(handle_file, "%s,is_cheat",file_head);
+        //PrintToChatAll("max_angle_tick: %d",g_iMaxAngleHistory);
+    }
+    bool first_meme2 = false;
+    char cheat_data[19999];
+    for (int i = 0; i < g_iMaxAngleHistory; i++)
+    {
+        if (idx == g_iMaxAngleHistory)
+        {
+            idx = 0;
+        }
+        if (IsVectorZero(g_fEyeAngles[client][idx]))
+        {
+            break;
+        }
+        // Nothing to compare on the first iteration.
+        if (i == 0)
+        {
+            vLastAngles = g_fEyeAngles[client][idx];
+            idx++;
+            continue;
+        }
+        vAngles = g_fEyeAngles[client][idx];
+        fAngleDiff = GetVectorDistance(vLastAngles, vAngles);
+        // If the difference is being reported higher than 180, get the 'real' value.
+        if (fAngleDiff > 180)
+        {
+            fAngleDiff = FloatAbs(fAngleDiff - 360);
+        }
+        
+        if (fAngleDiff > AIM_ANGLE_CHANGE)
+        {
+            PrintToChatAll("[CrowAI] %N 暴力自瞄",client);
+            break;
+        }
+        int mouse[2];
+        mouse = g_iMouse[client][idx];
+        if(first_meme2 == false){
+            first_meme2 = true;
+            Format(cheat_data, sizeof(cheat_data), "%f,%f,%f,%d,%d",g_fEyeAngles[client][idx][0],g_fEyeAngles[client][idx][1],fAngleDiff,mouse[0],mouse[1]);
+        }else{
+            Format(cheat_data, sizeof(cheat_data), "%s,%f,%f,%f,%d,%d",cheat_data,g_fEyeAngles[client][idx][0],g_fEyeAngles[client][idx][1],fAngleDiff,mouse[0],mouse[1]);
+        }
+        
+        vLastAngles = vAngles;
+        Last_AngleDiff = fAngleDiff;
+        idx++;
+    }
+    int is_cheat = GetConVarInt(g_cheat);
+    WriteFileLine(handle_file, "%s,%d",cheat_data,is_cheat);
+    //PrintToChatAll("[CrowAI] %N 分析完毕",client);
+    CloseHandle(handle_file);
+}
+
+void Aimbot_ClearAngles(int client)
+{
+    /* Clear angle history and reset the index. */
+    g_iEyeIndex[client] = 0;
+    
+    for (int i = 0; i < g_iMaxAngleHistory; i++)
+    {
+        ZeroVector(g_fEyeAngles[client][i]);
+        g_iMouse[client][i][0] = 0;
+        g_iMouse[client][i][1] = 0;
+    }
+}
diff --git a/test.py b/test.py
new file mode 100644
index 0000000..8ef48aa
--- /dev/null
+++ b/test.py
@@ -0,0 +1,107 @@
+import os
+import tensorflow as tf
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+import functools
+import csv
+test_file_path = [
+    r"I:\csgoserver\steamcmd\steamapps\common\Counter-Strike Global Offensive Beta - Dedicated Server\csgo\addons\sourcemod\测试数据1.csv"]
+
+train_file_path = [
+    r"I:\csgoserver\steamcmd\steamapps\common\Counter-Strike Global Offensive Beta - Dedicated Server\csgo\addons\sourcemod\训练数据.csv"]
+
+csv_data = []
+csv_data_x = []
+csv_data_y = []
+csv_data_all_num = []
+csv_data_y_avg = {}
+with open(train_file_path[0], 'r') as f:
+    reader = csv.reader(f)
+    for row in reader:
+        csv_data.append(row)
+csv_data_x = csv_data[0]
+for index in range(1, len(csv_data)):
+    for zeus in range(len(csv_data[index])):
+        if csv_data_x[zeus] == 'is_cheat':
+            continue
+        if csv_data_x[zeus] in csv_data_y_avg.keys():
+            csv_data_y_avg[csv_data_x[zeus]] = float(
+                csv_data_y_avg[csv_data_x[zeus]]) + float(csv_data[index][zeus])
+
+        else:
+            csv_data_y_avg[csv_data_x[zeus]] = float(csv_data[index][zeus])
+for key in csv_data_y_avg:
+    csv_data_y_avg[key] = float(csv_data_y_avg[key]) / float(len(csv_data) - 1)
+
+print(csv_data_y_avg)
+
+
+def process_continuous_data(mean, data):
+    # 标准化数据
+    data = tf.cast(data, tf.float32) * 1/(2*mean)
+    return tf.reshape(data, [-1, 1])
+
+
+def create_model():
+    numerical_columns = []
+    for feature in csv_data_y_avg.keys():
+        num_col = tf.feature_column.numeric_column(
+            feature, normalizer_fn=functools.partial(process_continuous_data, csv_data_y_avg[feature]))
+        numerical_columns.append(num_col)
+    preprocessing_layer = tf.keras.layers.DenseFeatures(numerical_columns)
+    model = tf.keras.models.Sequential([
+        preprocessing_layer,
+        tf.keras.layers.Dense(128, activation='relu'),
+        tf.keras.layers.Dense(128, activation='relu'),
+        tf.keras.layers.Dense(1, activation='sigmoid'),
+    ])
+
+    model.compile(optimizer='adam',
+                  loss=tf.losses.BinaryCrossentropy(
+                      from_logits=True),
+                  metrics=['accuracy'])
+
+    return model
+
+
+def get_dataset(file_path):
+    dataset = tf.data.experimental.make_csv_dataset(
+        file_path,
+        batch_size=32,
+        label_name='is_cheat',
+        na_value="?",
+        num_epochs=1,
+        ignore_errors=True)
+    return dataset
+
+
+def start_predict():
+    test_data = get_dataset(test_file_path)
+    print(test_data)
+    test_loss, test_accuracy = model.evaluate(test_data)
+    predict_data = model.predict(test_data)
+    print("anti-aimbot model, accuracy:{:5.2f}%".format(100 * test_accuracy))
+    num_cheat = 0
+    for pre, result in zip(predict_data[:20], list(test_data)[0][1][:20]):
+        num_access = 100 * pre[0]
+        if num_access >= 60:
+            num_cheat = num_cheat + 1
+        print("player is aimbot :{:5.2f}% ".format(
+            num_access), " is cheat: ", ("yes" if bool(result) else "no"))
+
+    print("result:", num_cheat)
+    if len(predict_data) < 20:
+        print("数据太少,无法辨别")
+    if num_cheat >= (len(predict_data) / 2) - 2:
+        print("此人微自瞄")
+    else:
+        print("此人不是微自瞄")
+
+
+model = create_model()
+model.load_weights('./save/model_weight')
+start_predict()
+while True:
+    input("按任意键开始预测")
+    start_predict()
diff --git a/train.py b/train.py
new file mode 100644
index 0000000..d49bb17
--- /dev/null
+++ b/train.py
@@ -0,0 +1,103 @@
+import tensorflow as tf
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+import functools
+import csv
+import time
+from tensorflow.keras.callbacks import TensorBoard
+
+
+LABEL_COLUMN = 'is_cheat'
+
+# "I:\csgoserver\steamcmd\steamapps\common\Counter-Strike Global Offensive Beta - Dedicated Server\csgo\addons\sourcemod\训练数据.csv"
+train_file_path = [
+    r"I:\csgoserver\steamcmd\steamapps\common\Counter-Strike Global Offensive Beta - Dedicated Server\csgo\addons\sourcemod\训练数据.csv"]
+test_file_path = [
+    r"I:\csgoserver\steamcmd\steamapps\common\Counter-Strike Global Offensive Beta - Dedicated Server\csgo\addons\sourcemod\测试数据1.csv"]
+csv_data = []
+csv_data_x = []
+csv_data_y = []
+csv_data_all_num = []
+csv_data_y_avg = {}
+with open(train_file_path[0], 'r') as f:
+    reader = csv.reader(f)
+    for row in reader:
+        csv_data.append(row)
+csv_data_x = csv_data[0]
+for index in range(1, len(csv_data)):
+    for zeus in range(len(csv_data[index])):
+        if csv_data_x[zeus] == 'is_cheat':
+            continue
+        if csv_data_x[zeus] in csv_data_y_avg.keys():
+            csv_data_y_avg[csv_data_x[zeus]] = float(
+                csv_data_y_avg[csv_data_x[zeus]]) + float(csv_data[index][zeus])
+
+        else:
+            csv_data_y_avg[csv_data_x[zeus]] = float(csv_data[index][zeus])
+for key in csv_data_y_avg:
+    csv_data_y_avg[key] = float(csv_data_y_avg[key]) / float(len(csv_data) - 1)
+
+print(csv_data_y_avg)
+
+
+def get_dataset(file_path):
+    dataset = tf.data.experimental.make_csv_dataset(
+        file_path,
+        batch_size=32,
+        label_name=LABEL_COLUMN,
+        na_value="?",
+        num_epochs=1,
+        ignore_errors=True)
+    return dataset
+
+
+def process_continuous_data(mean, data):
+    # 标准化数据
+    data = tf.cast(data, tf.float32) * 1/(2*mean)
+    return tf.reshape(data, [-1, 1])
+
+
+raw_train_data = get_dataset(train_file_path)
+raw_test_data = get_dataset(test_file_path)
+examples, labels = next(iter(raw_train_data))  # 第一个批次
+print("EXAMPLES: \n", examples, "\n")
+print("LABELS: \n", labels)
+
+numerical_columns = []
+for feature in csv_data_y_avg.keys():
+    num_col = tf.feature_column.numeric_column(
+        feature, normalizer_fn=functools.partial(process_continuous_data, csv_data_y_avg[feature]))
+    numerical_columns.append(num_col)
+
+preprocessing_layer = tf.keras.layers.DenseFeatures(numerical_columns)
+model = tf.keras.Sequential([
+    preprocessing_layer,
+    tf.keras.layers.Dense(128, activation='relu'),
+    tf.keras.layers.Dense(128, activation='relu'),
+    tf.keras.layers.Dense(1, activation='sigmoid'),
+])
+
+model_name = "anti_aimbot-{}".format(int(time.time()))
+TensorBoardcallback = TensorBoard(
+    log_dir='logs/{}'.format(model_name),
+    histogram_freq=1, batch_size=32,
+    write_graph=True, write_grads=False, write_images=True,
+    embeddings_freq=0, embeddings_layer_names=None,
+    embeddings_metadata=None, embeddings_data=None, update_freq=500
+)
+
+model.compile(
+    loss='binary_crossentropy',
+    optimizer='adam',
+    metrics=['accuracy'])
+train_data = raw_train_data.shuffle(1500)
+test_data = raw_test_data
+history = model.fit(train_data, epochs=85, callbacks=[TensorBoardcallback])
+test_loss, test_accuracy = model.evaluate(test_data)
+predict_data = model.predict(test_data)
+print('anti-aimbot: \n\nTest Loss {}, Test Accuracy {}'.format(test_loss, test_accuracy))
+for pre, result in zip(predict_data[:20], list(test_data)[0][1][:20]):
+    print("player is aimbot :{:5.2f}% ".format(
+        100 * pre[0]), " is cheat: ", ("yes" if bool(result) else "no"))
+model.save_weights('./save/model_weight')