⬅️ back

💵 Adult Income 💵 (2023)

An analysis of the classic Adult dataset where data cleaning and preprocessing techniques were applied, as well as modeling and deployment of a binary classifier. The following tools were used:

• Sklearn
• Streamlit
• FeatureEngine
• ImbalancedLearn
• Pandas
• Plotly

📈 Tomato market analysis 📈 (2023)

The objective of this project was to analyze the flow of tomato prices between national markets. The database was obtained through the SNIIM system (Mexico). This project completed the activities:

• Data minning
• Data cleaning
• Exploratory Data Analysis

Librarys and tools:

• Scrapy
• Sklearn
• Pandas
• Plotly
• SQL

🕺🏽Character Classifier 🕺🏽(2022)

The main task is to classify Simpsons characters using a convolutional neural network, the activities include:

• Data acquisition
• Exploratory analysis of the data set
• Preprocessing and manipulation
• Modeling
• Evaluation of the model.

Highlights:

• Preprocessing contemplates centering and filling operations since the exploratory analysis revealed images with different dimensions and aspect ratio.
• The transformed images were stored on disk using TFRecords to test their efficiency in manipulation and consumption in the training process.
• The convolutional network model was trained in TensorFlow using a custom class inherited from tf.module with weight initializers subject to a uniform lecun distribution.
• Sklearn was useful for model evaluation with metrics such as confusion matrix, precision, recall and f1.

🍔 Fast Food Classifier 🍔(2022)

In this project we have classified different fast foods and different representations (commercial, amateur photos, professional photos, etc). The activities include:

• Preprocessing and manipulation
• Modeling
• Model evaluation

Aspects to be highlighted:

• The pre-trained EfficientNet B0 model was chosen as the base model for the classifier.
• The cosine annealed decay and reset algorithm was used for the learning rate in order to obtain improvements in the neural network weight optimization process.

🌏 Research on World Models 🌏 (2021-2023)

Master thesis that improves the performance of world-model based agents by reducing the number of parameters in appropriate models such as PCA and VAE. The project was developed in Python and Tensorflow.

Aspects to be highlighted:

• To facilitate the inference of the proposed models and experimentation, a local deployment with serving_default of TensorFlow was chosen.
• Weights & Biases was used to track experiments and model performance.
• An optimization based on evolutionary algorithms was chosen instead of gradient-based algorithms.

🤖 Agricultural robot for spraying 🤖 (2020)

Agricultural robot prototype for monitoring and spraying tasks in tomato greenhouses. The project covered the mechanical design and development of control systems at the functional prototype level. Solidworks, Matlab and Labview were used.

Aspects to be highlighted:

• The robot was designed from scratch and validated with specialized software to withstand the real conditions of agricultural environments.
• The control system was based on classical algorithms such as PID and PD for motor control by means of system identification and controller design.
• The power electronics were chosen to be modular in order to facilitate repair or scaling up.
• The development boards were programmed in C++ language.