MQL5 Algo Trading’s Post

Ever wondered how modern vector databases like FAISS, Milvus, or Pinecone can search millions of embeddings in milliseconds? The secret lies in a powerful algorithm called HNSW (Hierarchical Navigable Small World) one of the most efficient graph-based methods for approximate nearest neighbour (ANN) search. HNSW is a combination of skip lists and NSW (Navigable Small World). The hierarchical feature comes from skip lists which stores data points in layers with all data points in the bottom most layer and least data points in the top most layer and NSW part connects the data points in a layer to each other and with the lower layers. During search, the starting point is pre defined and moves to lower layers if the distance of the query vector is smaller to a lower layer than the current node and it stops when there is no node having smaller distance than the current one. HNSW has become the backbone of most modern vector search systems balancing speed, scalability, and accuracy. It’s fascinating how a graph-based approach can make high-dimensional search so efficient. image source: https://lnkd.in/gHsrAxbh

Dive into advanced techniques for improving MetaTrader 5 libraries with this insightful article focusing on graphical object events. Discover how you can define precise changes in properties and track the history of modifications, enabling enhanced analytical tools with memory capabilities. Learn about handling complex scenarios, such as multiple object additions and chart-linked object restoration. Benefit from structured methods for capturing object renaming sequences and managing removed graphical objects efficiently. Perfect for developers seeking to refine trading tools, this article offers detailed, practical approaches for leveraging the full power of MQL5's dynamic arrays and chart management classes. #MQL5 #MT5 #AlgoTrading #Indicator https://lnkd.in/dCcrifJx

🚀 Is Your VectorDB Running at Its Best? Working with millions of embeddings? The real bottleneck might not be your model — it’s your index. In my latest Medium post, I break down how to fine-tune HNSW for speed, recall, and memory efficiency: 🔑 Key parameters: M, ef_construction, ef 🧪 Step-by-step tuning & evaluation 📈 Tips to keep performance strong as your dataset grows 👉Read the full guide here: https://lnkd.in/ed54wdgS \ #vector_database #semantic_search #llm_as_judge #recall #fine_tuning #pinecone #weaviate #elastic_search

🚀 Problem Solving Update | GFG + LeetCode Daily Challenges Today’s session was all about reinforcing concepts and tackling tricky logic! 💪 🔹 GFG POTD – Bottom View of Binary Tree Had solved it earlier, revisited it today to strengthen understanding. 📘 Approach Summary: Use a map to store the bottommost node for each horizontal distance (HD). Perform level-order traversal (BFS) with a queue holding (node, HD). For each node, update the map so that the latest node at that HD is stored (since BFS ensures bottom-most replacement). Finally, extract the node values from the map (sorted by HD) to get the bottom view. This approach beautifully combines BFS traversal with mapping of node levels — a neat and efficient way to visualize the tree from below 🌳 🔹 LeetCode Daily Challenge – Avoid Flood in The City This one was a logic-heavy simulation problem! 📘 Approach Summary: Maintain a map to track the last day each lake was filled. Maintain a sorted set of available dry days (indices where rains[i] == 0). For each day: If it rains on a lake already full → find the next available dry day after its last fill day and assign it to dry this lake. If no such dry day exists → 🚨 flood occurs. Otherwise, update records accordingly. Unused dry days can dry any lake at the end. Efficiently combining maps + sets ensures we handle all scenarios without brute-forcing every combination ⚙️ #ProblemSolving #LeetCode #GeeksforGeeks #DSA

FactRAG and HybridRAG insert a KG‑aware retriever between the index and the generator. Logged traversal paths give auditability. In regulated domains (e.g. finance), the graph acts as a guardrail. Gains scale with KG cleanliness and retriever quality. HybridRAG = vector recall plus graph constraints. Later work, like PolyG, optimises traversal. Results on FinanceBench: • 6% fewer hallucinations • 80% fewer tokens • Complexity drops from O(n^2) to O(k·n) https://lnkd.in/dDT7ug7S https://lnkd.in/dfG2qzGM

Data quality is subjective... until you sit with the user. I once assumed casing data had to be “complete” to be valuable. But two geotechs set me straight: “We just need to know casings *exist.* Details come later.” The data didn’t need to be perfect, it just had to support the way they worked. ODIN by PandoScape Inc. continues to evolve with this in mind. No fixed rules. No one-size-fits-all. Your interpreters define the standard. We make sure your subsurface data meets it. Learn more here: https://lnkd.in/gzwD4t-Z

A single missed "manual" step can drift into shareholder reports, regulatory reports, and entire interpretations. The stories I've heard from geotechs would give an exploration VP nightmares. Once bad data (or missing data) is baked in, every map, model, and approval carries that same bad value. And fixing it means reworking everything downstream. That’s why ODIN by PandoScape Inc. keeps interpreters on one source of truth: → Automated ingestion from every source → QC that runs nonstop → Instant flags before errors spread into schedules or budgets If you’ve chased a bad data halfway through a project, you know why we built this. Your org pays millions to create the foundations for a data playground. ODIN keeps the playground safe for your interpreters to play in.

Data quality is subjective... until you sit with the user. I once assumed casing data had to be “complete” to be valuable. But two geotechs set me straight: “We just need to know casings *exist.* Details come later.” The data didn’t need to be perfect, it just had to support the way they worked. ODIN by PandoScape Inc. continues to evolve with this in mind. No fixed rules. No one-size-fits-all. Your interpreters define the standard. We make sure your subsurface data meets it. Learn more here: https://lnkd.in/g2hMBMuk

One of the most powerful new capabilities in Delta Lake 4.0 is support for collations, giving you much finer control over how text is compared and sorted. ✅ In this clip, Youssef Mrini explains how to enable collations: 🔹 Define a default collation at the table level 🔹 Customize collations for individual columns 🔹 Use ALTER TABLE or ALTER COLUMN to add them to existing tables Whether you’re handling multilingual datasets or need case-insensitive searches, collations make your queries more accurate and flexible — with minimal changes to your code. 🎥 Watch the full webinar to explore more: https://lnkd.in/eKgJ3tXM #opensource #oss #deltalake #linuxfoundation Scott Haines

Kendall's Tau - Quick Introduction How to compute correlations among ordinal variables? We can't use Pearson correlations here but suitable alternatives are -Spearman rank correlations and -Kendall's Tau. Most analysts are familiar with Spearman correlations but we feel Kendall's tau is the preferable measure. So what is it and why is it better? For a quick introduction, read up on https://lnkd.in/eGScyRJu Hope it helps! #statistics #spss #dataanalysis