numpy

150 problems

• Accuracy scoreEasy
• AdaBoost fit (decision stumps)Medium
• Adadelta optimizerMedium
• Adagrad optimizerEasy
• Adam optimiser stepMedium
• Adamax optimizerMedium
• AdamW step (decoupled weight decay)Medium
• Affine coupling layer (normalizing flow)Medium
• Autocorrelation functionMedium
• Autoencoder forward passEasy
• Bag-of-words encodingEasy
• BatchNorm forward (train + eval modes)Medium
• Beam search decodingMedium
• Bellman equation for value iterationMedium
• Bernoulli Naive Bayes classifierMedium
• Best Gini-based split (decision tree)Medium
• Binary classification with logistic regressionEasy
• Conv2D forward (naive, stride 1, no pad)Medium
• Conv2D forward (padding + stride)Medium
• Max pooling 2D forwardEasy
• Average pooling 2D forwardEasy
• Conv2D backward (dx, dW)Hard
• Mini-CNN forward (capstone)Medium
• Token embedding lookupEasy
• Sinusoidal positional encodingEasy
• Scaled dot-product self-attentionMedium
• Causal mask: build + applyEasy
• Multi-head split + combineMedium
• Multi-Head Attention (full layer)Medium
• LayerNorm forwardEasy
• Transformer block forward (pre-LN, residual)Hard
• Linear forward (Wx + b)Easy
• ReLU forward + backwardEasy
• Sigmoid forward + backwardEasy
• MSE loss + backwardEasy
• Linear backward (chain rule)Medium
• SGD step (in place)Easy
• Train a 2-layer MLP on XORHard
• Cohen's kappa scoreMedium
• Confusion matrixEasy
• Contrastive lossMedium
• Warmup + cosine decay LR scheduleEasy
• Covariance matrixEasy
• Cross-entropy gradientMedium
• DBSCAN clusteringMedium
• Dense block with 2D convolutionsMedium
• Dice lossEasy
• Diffusion forward processMedium
• Divide dataset by feature thresholdEasy
• DPO lossMedium
• Dropout layer (forward & backward)Medium
• Dynamic-Tanh (DyT)Medium
• Efficient sparse window attentionMedium
• Elastic-Net regression (gradient descent)Medium
• Euclidean distance matrixEasy
• Exponential moving average of weightsEasy
• F1 score (binary classification)Easy
• Fisher-Yates shuffleEasy
• FlashAttention tiled forwardHard
• Focal loss (multiclass)Medium
• Gaussian process regression (RBF)Hard
• GCN layer (message passing)Medium
• GELU backward (tanh approximation)Medium
• GELU forward (tanh approximation)Easy
• Generate random subsets (bootstrap)Easy
• GMM E-step (responsibilities)Medium
• Gradient checkpointing forwardEasy
• Clip gradients by global L2 normEasy
• Graph LaplacianMedium
• Greedy decoding loopEasy
• Group normalizationMedium
• Grouped-query attentionMedium
• GRPO objectiveHard
• He weight initializationEasy
• Huber & Hinge lossesEasy
• Instance normalizationMedium
• Inverse-transform samplingEasy
• Jaccard similarityEasy
• k-fold cross-validationEasy
• KL divergence (discrete)Easy
• K-means: one iterationMedium
• k-NN classification (majority vote)Easy
• KV cache for autoregressive inferenceMedium
• KV cache compression (MLA)Hard
• Label-smoothed cross-entropyMedium
• Leaky ReLU activationEasy
• Learning-rate range testMedium
• Linear regression — gradient descentEasy
• Linear regression — normal equationMedium
• Lion optimizer stepMedium
• Log-softmaxEasy
• Logistic regression — gradient descentMedium
• LoRA forward passEasy
• Matmul backwardMedium
• Multi-class cross-entropy lossEasy
• Mutual informationMedium
• Nesterov accelerated gradientMedium
• Neural ODE forward EulerMedium
• Noisy top-k gatingMedium
• Orthonormal basis (Gram-Schmidt)Medium
• PageRank (power iteration)Medium
• Pairwise cosine-similarity matrixEasy
• Pearson correlation coefficientEasy
• Pegasos kernel SVMMedium
• Perplexity from log-probsEasy
• Pointwise mutual information (PMI)Medium
• Policy gradient with REINFORCEMedium
• PPO clipped objectiveMedium
• Precision@k and NDCG@kMedium
• Precision metricEasy
• Principal Component Analysis (PCA)Medium
• Prioritized experience replayMedium
• Q-learning for MDPsMedium
• Rejection samplingMedium
• Repetition penalty (HuggingFace-style)Medium
• Reservoir samplingMedium
• Residual block with shortcutEasy
• Ridge regression lossEasy
• RMSNormEasy
• RMSprop optimizerEasy
• Vanilla RNN cell forwardMedium
• ROC-AUC from scratchMedium
• Rotary Position Embedding (RoPE)Hard
• Scaled dot-product attentionMedium
• SELU activationEasy
• Sequence padding & maskingEasy
• SGD with momentum (one step)Easy
• Shannon entropyEasy
• Singular Value Decomposition (2x2)Hard
• Softmax backwardMedium
• Softmax from scratchEasy
• Softmax (multinomial) regressionMedium
• Sorted polynomial featuresMedium
• Sparse mixture-of-experts layerHard
• Speculative decoding verificationMedium
• SwiGLU activationMedium
• Swish / SiLU activationEasy
• TD(0) value updateEasy
• Temperature scalingEasy
• TF-IDFMedium
• Time-series anomaly detectionMedium
• Top-k samplingEasy
• Top-p (nucleus) samplingMedium
• Triplet lossMedium
• UCB1 multi-armed banditEasy
• VAE ELBO lossMedium
• Viterbi algorithmMedium
• Wasserstein distance (1-D)Medium
• Weighted cross-entropyEasy
• Weighted multinomial samplingEasy